--- title: Software changelog description: The main changes in each Teloa version and its actual release status. lastVerified: 2026-11-06 outline: 2 --- # 8.2 Software changelog ## 0.2.31 (stable) 2026-11-06 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.31) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.31) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) - **Directory menus display fully**: fixes clipped panels for new conversations, conversation filters, folder actions, tasks, automations and execution history. - **Complete voice playback**: empty speech recognition no longer interrupts and truncates a reply. Meaningful interruptions, mute and hang-up remain available. - **Window adaptation**: panels follow resizing and scrolling, open upward near the window edge, and scroll long content within the panel. - **Keyboard and context**: preserves Escape, outside dismissal, focus return and selected filters. Upgrades keep data version 18. - **Voice follow-ups**: a new question received during generation continues in the original session’s next turn, preserving the current turn’s admission record and avoiding continuation checkpoint conflicts. ## 0.2.27 (Stable) 2026-11-05 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.27) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.27) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) · [Release docs](https://docs.teloa.ai/v/0.2.27/en/start/quickstart) - **Compact workbench:** retains the approved interface with consistent typography, controls and list density, personal font scaling and keyboard operation. - **Group discussion and formal work:** ordinary replies do not create an extra Task/Run. Explicitly assigned work keeps its task, progress and deliverables. Completed formal work stays reachable; ordinary discussion does not show a fictitious work entry. - **Materials and continuity:** reading comes first, with file information available on demand. Back navigation and closing details preserve the original object and unsent draft, avoiding duplicate creation or accidental sending. - **Business scope:** a small lock explains fixed scope while retaining the original business ownership and explicit grants. - **Maintenance:** data version is **18**, with direct compatibility `[18]`. Keep the original home, identity, database, sessions and workspace, and migrate through normal backup and upgrade. Rollback requires the matching old program and pre-upgrade backup. ## 0.2.12 (Stable) 2026-11-05 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.12) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.12) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) - **Natural-language work:** employee and task creation start with Markdown, carrying business and source context. Drafting, owner confirmation and execution remain separate. Employee mentions and business context are checked against current identity. - **Readable deliverables and materials:** result bodies come first, with source, files and editing opened as needed. Multi-format previews preserve original files and explicit versions. Citations open real sources; historical references are not presented as current versions. - **Permissions:** a concise entry shows business, materials, tools and ongoing work. Read, write and edit paths can be configured separately and checked against frozen run grants. Rejection or cancellation is not permission to change arguments or seek reauthorization. - **Continuity:** eligible tasks recover with their original Task/Run under checkpoint, permission and host checks; ordinary conversations offer manual continuation when available. Unknown external effects are not replayed, and version conflicts preserve drafts for review. - **Materials and memory:** verify actual retrieval scope, material versions, index readiness and memory settings. General retrieval without isolation is not project-only visibility. Review model conclusions against their sources. - **Maintenance:** normal upgrade continuation handles historical completed operations while preserving the original home, identity, tasks, sessions and workspace. Data version is **17**, with direct compatibility `[17]`; supported older data follows the normal backup/migration path. Rollback needs the old program and a matching pre-upgrade backup. Native installation was verified on macOS ARM64 with Node.js 24.22.1. Docker targets Linux ARM64. Other models, devices and platforms retain their respective configuration and support boundaries. See [installation](https://docs.teloa.ai/v/0.2.31/markdown/en/start/quickstart.md) and [0.2.12 walkthrough](https://docs.teloa.ai/v/0.2.31/markdown/en/start/whats-new.md). ## 0.2.10 (Stable) 2026-10-24 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.10) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.10) - Patch release: the native package manifest now records the pinned DSH version (`dshVersion`), matching the manifest schema. Features are unchanged from 0.2.9. - Docker Hub now serves `teloa/teloa:0.2.10` (Linux ARM64, digest `sha256:4f91a98b…b2fd`). Install and upgrade: GitHub Release `teloa-cli-0.2.10.tgz`, npm `@teloa/cli@0.2.10`, Docker `teloa/teloa:0.2.10` (Linux ARM64, Compose at `https://docs.teloa.ai/downloads/0.2.10/compose.yaml`). Native upgrades from 0.2.9 to 0.2.10 keep tasks and identity. ## 0.2.9 (Stable) 2026-10-24 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.9) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.9) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) - Next-step suggestions in the composer: after a reply, an empty composer shows one grey model-written sentence as the likely next input. Tab puts it into the draft and never sends; typing hides it and clearing the input brings it back; Settings > General can turn it off. Copy is available in Simplified Chinese, Traditional Chinese and English. - Suggestions are produced through the host's admitted auxiliary-model path; no provider HTTP is called directly, and no message, task or execution is created. When the model is unavailable the UI falls back or shows nothing. - Group chats keep the online-1.0 unified left-author layout (the owner's own messages are flat and bubble-less as well). Install and upgrade: GitHub Release `teloa-cli-0.2.9.tgz`, npm `@teloa/cli@0.2.9`, Docker `teloa/teloa:0.2.9` (Linux ARM64). Native upgrades from 0.2.8 to 0.2.9 keep tasks and identity. ## 0.2.8 (Stable) 2026-10-23 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.8) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.8) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) - The employee API accepts a per-request permission mode and custom model provider; fixed results automatically create a formal artifact, complete the linked task and update work progress. - Employee-package materials support body preview and role-scoped chat retrieval; the library can add all of an employee's materials to local retrieval in one action. Employee-package updates provide an impact preview and transactional switch; unload keeps employees, materials, tasks and historical deliverables while detaching connection/tool/extension instances. - The market creation menu now includes Create skill and Create app; hard uninstall for installed skills carries over from 0.2.7. - The Teloa standard preset is grouped under built-in Agent presets; PTC code modes have clear names and accurate descriptions. - Settings groups have dividers and hierarchy; the non-Chinese sidebar brand no longer overlaps the new/search actions. - Node.js security baseline is 24.22.1. SDK remains compatible with `@teloa/sdk@0.2.4`. Install and upgrade: GitHub Release `teloa-cli-0.2.8.tgz`, npm `@teloa/cli@0.2.8`, Docker `teloa/teloa:0.2.8` (Linux ARM64; Compose at `https://docs.teloa.ai/downloads/0.2.8/compose.yaml`). The native 0.2.7 to 0.2.8 upgrade preserves tasks and identity. ## 0.2.7 (Stable) 2026-10-23 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.7) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.7) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) - Added real hard uninstall for installed skills from the workbench and natural language, with explicit host confirmation and protection for current selections, loaded packages and historical run evidence. - Closed production-QA gaps in task/group execution gates, AutoDream, project and employee memory, creation-draft recovery, installed-market management and directory focus. - Verified the 0.2.6 to 0.2.7 native CLI upgrade keeps tasks and data, and the 0.2.7 Docker image starts on a 0.2.6 data volume with tasks and employees preserved. - Public source CI passed; fresh npm install, 13 lifecycle checks and four-page UI acceptance passed. SDK remains compatible with `@teloa/sdk@0.2.4`. ## 0.2.6 (stable) 2026-10-22 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.6) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.6) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) - Fix original SKILL license reading and sharing, original employee-draft confirmation and missing-template-field diagnostics. - Check actual skill availability and report same-name source conflicts accurately. - Fix successive sharing and imports while retaining unfinished requests and choices. - Clarify group-routing and delivery-review parameters; verify draft, review and final delivery with real MiniMax. - Complete natural-language creation, use, sharing and receiving instructions for AI employees, AI teams, Apps and SKILL, with eight Chinese and English interface screenshots. - Pass clean macOS ARM64 native installation, 13 lifecycle checks and four UI pages, plus Linux ARM64 empty-volume startup and task persistence across restarts. SDK compatibility remains `@teloa/sdk@0.2.4`. ## 0.2.5 (stable) 2026-10-22 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.5) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) · not published to npm **Historical channel note:** 0.2.5 was not published to npm; npm publishing resumed with 0.2.6, and the current release is 0.2.8. - Fix employee execution preparation when optional browser or computer providers are unavailable; tool availability and authorization still apply. - Real MiniMax verification covers API, SDK and MCP invocation, fixed results, continuation and cancellation; `@teloa/sdk@0.2.4` remains compatible. - Align Docker runtime version reporting and generated-carrier ownership; empty-volume startup and task persistence across restarts passed. - Organize the developer guide around AI employees, AI teams, Apps and SKILL, with runnable downloads. ## 0.2.4 (stable) 2026-10-21 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.4) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.4) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) This update makes AI employees and AI teams more dependable for continuous business work. It builds on the unified results sidebar and collaboration interface delivered in 0.2.3. ### Workbench and first use - Clearer navigation and page hierarchy for the workbench, conversations, projects, employees, business areas, library, and market. Your avatar opens My twin, while Settings keeps its own entry. - Personal and employee conversations keep their editors and separate text drafts when first opened, switched, or refreshed. - Superseded static product rules from earlier conversation context are excluded from subsequent model inputs; original conversation records, role responsibilities, and permission boundaries are preserved. - When the old default lacks a key and exactly one BYOK provider is configured, Teloa initializes the default model. Valid defaults and explicit model selections are preserved. - Group routing follows the current default model and its reasoning configuration. Authentication errors are distinguished from normally completed empty replies; an empty reply is retried at most once. - Natural-language task creation prepares a draft first. If no task was actually created, an inline notice offers Create with a form with the original goal. Saving does not run it. Explicit assignments use the native approval card: approval starts execution, while rejection creates no task. - Archived conversations can be restored with Undo. Market loading, retries, and cancellation have clearer states, and high-impact actions such as retirement require confirmation appropriate to their effects. - Create projects directly from the project overview with a Markdown description and preserved drafts. Business project filters survive tab navigation and stay isolated by business and owner connection. - General work shows applicable menu actions. Reference lists, documents, and the return path remain reachable in narrow windows. ### Installation and template compatibility - Source setup recognizes another Docker “volume not found” response and creates the database volume. Permission and connection errors still stop setup. - Market employee previews and the host share template validation, preserving structured responsibilities and runtime presets. Model accounts from a template are not adopted on installation. The marketplace website also shows triggers, autonomous actions, confirmation points, escalation rules and delivery checks. - Browse AI employees, AI teams, Apps and Skills as four independent categories. Skills can be filtered and added separately; Apps keeps its existing skills and connections. - Fix startup after enabling the official web search option in Settings. - Resuming an app import fully validates and reuses the owner's existing pinned-version skills, downloading only missing sources. Corrupt content or a source/version mismatch stops the import explicitly and keeps the original request available for review. - Native Auto review requests are associated with the current tool and work, avoiding an interruption caused by missing admission context. Revocation, cancellation, and work budgets are still checked per request. ### Employee APIs and integrations - Business systems can use the stable HTTP API, OpenAPI 3.1 contract, and TypeScript / JavaScript SDK to invoke an enabled employee, read status and versioned results, continue the same work, or request cancellation. - External agents can invoke the same work capabilities through Streamable HTTP at `/mcp`. Create keys in Settings > Employee integrations, binding each to one employee version, business scope, permissions, expiry, and concurrency limit. Keys support rotation and revocation. - Optional signed CloudEvents callbacks keep durable delivery records, retry failures automatically, and support manual redelivery. Verify signatures and deduplicate by event ID. Execution success and business acceptance are recorded separately. - See [Employee API, SDK and MCP](https://docs.teloa.ai/v/0.2.31/markdown/en/develop/employee-api.md) for setup, errors, and examples. This release does not claim A2A or MCP Tasks extension compatibility. ### Custom MCP apps and permissions - Describe an app in Markdown, then enter its endpoint and multiple authentication headers on the draft confirmation card. Validate connections, retry failures, inspect, disconnect, or delete them; credentials are stored encrypted. - Owners can explicitly select Full access while the default permissions remain unchanged. Already authorized MCP tools do not require an extra connection approval in that mode; employee tools and business authorization still apply independently. - Tools explicitly marked read-only, non-destructive, and not requiring MCP Tasks can run with bounded concurrency in the native loop. Write tools and those with missing or conflicting annotations remain sequential. ### Business sources drive employee work - New events from a connected business source (such as a log app) can start the matching employee's work one by one, keeping the link between source, object and task. - The business home shows which sources are actually connected, based on real sync rules and connection status. While reading it shows "to be checked" instead of treating a mapping alone as connected. - Advanced business settings explain when no sync configuration exists and where to describe the requirements. An empty configuration offers no enable or sync-now action. - Task goals and results use shared typography, and the browser tab title follows the current page. ### More reliable group handoffs and continuations - When an employee hands work to another employee in a group, the handoff starts the work with the tools already granted for that run. A failed handoff route shows its status under the message and can be rearranged by the owner. - Quoting an employee reply, or your own original assignment message, to send work back continues the original task instead of creating a same-titled task. Each routing turn is isolated from earlier turns. - When the original task is paused, the continuation explains why and offers to resume the original work in place after confirmation. - Retrying a message after a stop now settles instead of staying in "arranging". ### Steadier delivery review - When a review receipt has the wrong shape or citations, the hint names exactly what is wrong (field, citable passage range, block numbers) so the model can correct it once. - Automatic context compaction during review no longer swallows review content, and compaction can run more than once in a turn. - A refused call to a tool the model was never given no longer fails the whole delivery, and URL paths in review text are no longer blocked as file reads. - Provider overload responses (such as 529) are retried. ### Work budgets and recovery - Work budgets count model time in a rolling 24-hour window and explain why work stopped and what to do next; group messages explain budget rejections too. - Run starts are admitted atomically and queued plans are kept durably. - After an explicit `teloa upgrade`, work that was paused only by the restart resumes automatically; work you paused yourself or with operations still to verify stays paused. ### Install and upgrade Use the native package attached to this release or the matching npm / Docker Hub version. Native package integrity is verified on macOS ARM64; Docker publication targets Linux ARM64. Keep the original home and data, and use the official CLI backup and upgrade flow. Do not initialize the existing database again. See the [installation guide](https://docs.teloa.ai/v/0.2.31/markdown/en/start/quickstart.md) and [full changelog](https://docs.teloa.ai/v/0.2.31/markdown/en/changelog/index.md). ## 0.2.3 (stable) 2026-10-19 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.3) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.3) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) This update improves deliverable review and everyday collaboration. It builds on the AI employees, AI teams, projects and project memory, local document indexing, Codex / Claude Code harness integrations, and voice / TTS capabilities delivered in 0.2.2. ### One sidebar for results - Conversations, groups, tasks and projects use the same results sidebar. Select a result from the list, read it in place, and return with “All results”; there is no nested permanent directory. - Versions, sources, downloads, unsent feedback and the original working context remain available, with resizable and expanded reading views. - Group results show only versions explicitly shared with that group. Private versions and drafts are not automatically included. - Open saved results through the conversation directory’s “… / Work results” menu. An explicit title search can also find an original conversation with saved results but no chat messages, without sending a new message. ### Clearer collaboration controls - Employee collaboration invitations use a compact selection list while preserving members, scope, drafts and explicit confirmation. - Marketplace, team materials, details and related drawers share consistent spacing and scrolling, with less repetition and empty space. ### Reliable task listings Tasks with verified completed group replies remain readable. Backend, execution-host and client validation now agree, while ordinary attention requirements, invalid-state checks and ownership checks remain enforced. ### Install and upgrade Use the native package attached to this release or the matching npm / Docker Hub version. Native package integrity is verified on macOS ARM64; Docker publication targets Linux ARM64. Keep the original home and data, and use the official CLI backup and upgrade flow. Do not initialize the existing database again. See the [installation guide](https://docs.teloa.ai/v/0.2.31/markdown/en/start/quickstart.md) and [full changelog](https://docs.teloa.ai/v/0.2.31/markdown/en/changelog/index.md). ## 0.2.2 (stable) 2026-10-18 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.2) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.2) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) Today, **Teloa Community edition (0.2.2) is officially available.** Teloa is an AI-native team workspace. Describe a goal in natural language, create AI employees with ongoing responsibilities, and organize a team around projects, materials, memory and deliverables. This release brings together AI employees, AI teams, multiple execution harnesses, projects and project memory, local document indexing, voice interaction and versioned delivery. It also turns lessons from product development into two standardization initiatives made publicly available to the industry. ### 30-second workspace tour
Recorded in Community 0.2.2 with sample data. Download the 30-second tour
### AI employees: persistent roles for ongoing work Describe the role, working requirements and expected deliverables in natural language or Markdown, then review the proposal to create an AI employee. For example: > Create a product researcher to organize user feedback, check product assumptions, and deliver source-backed analysis and validation suggestions. List open questions when evidence is insufficient, and ask for my confirmation before publishing externally. Each employee can have its own responsibilities, working methods, business scope, reference materials, skills and available tools. These settings stay with the role so it can keep handling similar work. Chat privately with an employee, inspect its tasks, deliverables and accumulated experience, or add it to a team to collaborate on a shared goal. An employee's role identity is managed separately from the model. Changing models or starting a new task can preserve the same employee's responsibilities and work history. ### AI teams: organize responsibilities around shared results Combine multiple AI employees into a team with clear responsibilities for research, analysis, production and review. The marketplace provides team templates for product, marketing, sales and other work. They organize roles, skills, reference materials, optional connections and work templates together, reducing repetitive setup. After adding a template, review the members, target business and work arrangements before enabling the capabilities you need. Use `@` mentions in group chats to involve specific employees while retaining the source of assignments, collaboration and result replies. Main messages and topics are organized separately, with distinct entry points for ordinary chats, employee conversations and groups. **Persistent teams carry ongoing responsibilities; temporary execution teams serve the current task.** Independent parts can run in parallel before the responsible worker consolidates the results. Collaboration across business areas continues to follow each member's actual access to materials and tools. ### Multiple harnesses: DSH, Codex and Claude Code Teloa connects different execution engines to the same workflow: - **DSH** handles the main conversation, native execution and tool calls. - **Codex** handles independent code, file and tool tasks through its native execution provider. - **Claude Code / Claude Agent** connects to the native Claude execution path, handles independent work and returns the results. You can ask directly: > Use Codex to review the code this time, and have Claude Agent analyze these files. The results return to the original task, retaining the work's source and attribution. Model connections and harnesses are configured separately. Each execution provider uses its own authentication, and external engines require confirmation of their native working directory and permissions before launch. Choose the execution approach that fits the work, then follow its actual status and deliverables in the workspace. ### Projects and project memory: carry context into subsequent work Projects bring together goals, briefs, tasks, employees, groups, materials, automations and deliverables, with dedicated project conversations. Project goals, active rules, relevant materials and memories can participate in subsequent model requests. Inspect what was used and its sources to understand the basis of an employee's work. Review memory candidates, trace their sources, withdraw future use, and share a fixed revision. New conversations can choose whether to read and write long-term memory, and the owner decides whether personal preferences participate in a project. From the first discussion through later deliveries, a project can accumulate reusable background, rules and reasoning. ### Local document indexing and local models: put your own materials to work Add explicitly selected local Markdown files to an index, track material versions, and inspect citations and the locations of source excerpts during retrieval. Prepare retrieval models after confirmation on the materials page, with options including **Qwen3 Embedding and EmbeddingGemma**. Chat models handle understanding and generation; embedding models handle retrieval. The two are managed separately. Through **Ollama**, check the local chat environment, connect existing models, and confirm a download or cancel model preparation. The document index runs on the device or container where you deploy Teloa. Answers can use local models or remote models you configure. When you choose a remote chat service, relevant retrieved excerpts may be sent to that service as part of the current context. ### Natural-language tasks and versioned deliverables: from a brief to a result The main task-creation flow centers on a natural-language or Markdown brief. Describe the goal, requirements and delivery format; the source conversation, project and business are carried in from context. Follow execution, handle required approvals, stop work, preview files and collect deliverables in the same workflow. Even before creating employees, you can work independently with Teloa, the squirrel assistant, or the general-purpose Agent. Preview and download deliverables, then request revisions to a selection or the whole document. Each revision retains a clear version. Earlier results such as v1 and v2 remain available with their original attribution, and acceptance can refer to the specific version adopted at the time. The workspace previews Markdown, code, images, PDF, HTML, Office files and spreadsheets so you can inspect the actual delivery. ### Business workflows, dashboards and the marketplace Create or adjust business configurations in natural language, review and adopt the draft, then import records or connect authorized data sources. Business content and statistical dashboards are organized separately, so you can inspect individual records and observe changes across the business. Tasks, materials and deliverables can continue within the same business context. Discover AI employees, AI teams, skills, apps and connectors through dedicated marketplace entries. The apps page manages installed capabilities and connections. For more complex work, use step-by-step Workflow orchestration, PTC tool composition, and Ralph iteration when explicitly requested by the owner. Optional capabilities such as Browser Use, Computer Use and IM channels are configured and authorized separately as needed. ### Voice input, TTS and calls: collaborate by speaking Teloa brings voice input, recognition, speech output and calls into the workspace: - **Voice input:** SenseVoice supports local transcription. Recognized text enters an editable draft for the owner to review and send. - **Call recognition:** choose MiniMax cloud recognition or local SenseVoice. - **Text-to-speech:** MiniMax, OpenAI and ElevenLabs are supported, with separate model, voice and credential settings. - **Voices and previews:** MiniMax offers Mandarin and English voice selection and previews. Other providers use their corresponding model and voice settings. - **Continuous speech and call controls:** MiniMax supports continuous audio output, reducing repeated starts between short phrases, while retaining mute, interruption and hang-up controls. Call an employee or the squirrel assistant, or leave a voice message to add work requirements. Calls remain associated with the original conversation, and you can continue reviewing work and deliverables after a call ends. ### Auto Dream, employee experience and personal digital twins Auto Dream organizes private daily summaries from work evidence and records employee experience with its source and scope. You can inspect it and withdraw its use. Employee experience, project memory and personal digital twin materials are managed separately. A twin can accumulate judgment examples supplied by its owner and observations of working habits to help organize personal working methods. Each material retains its own scope of use. Methods you have confirmed as useful can also become skills or recurring-plan drafts. Work arrangements are explicitly enabled by the owner, allowing recurring work to develop into processes that can be inspected and adjusted. ### Two open standards: how AI works and acts safely To describe how AI workers are organized and where their security boundaries lie, **Teloa has made two standardization initiatives publicly available to the industry**, for developers, enterprise users and other platforms to implement, review and improve together. #### AI Worker Standard 1.0 Initiated and maintained by Teloa, **AI Worker Standard 1.0** provides working definitions and collaboration contracts for AI Workers, AI Teams and AI Organizations. It describes: - Worker identity, responsibilities and definition versions that persist across tasks. - Inputs, outputs, dependencies, authorization conditions and acceptance requirements. - Responsibilities, handoffs and shared deliverables between team members. - Organizational accountability, resource constraints and change governance. The standard distinguishes role definitions, actual employee instances and individual executions. It includes a core specification, a work and collaboration profile, and JSON and YAML examples to help different implementations express what a worker is responsible for, how it collaborates and how its deliverables are checked. [Read AI Worker Standard 1.0](https://github.com/teloa-ai/ai-worker-standard) #### Enterprise Agent Identity and Authorization Security Baseline Initiated and maintained by **Teloa AI Security**, the **Enterprise Agent Identity and Authorization Security Baseline v1.0** contains **240 verifiable requirements across 10 security domains**. The baseline covers identity authentication, delegation and permissions, authorization decisions, credential management, execution isolation, context security, output visibility, data and memory, lifecycle governance, and audit and response. It provides **L1 baseline, L2 standard and L3 enhanced** assessment levels, together with online and offline self-assessment tools, maturity reports and Excel assessment sheets for platform design, supplier evaluation and security review. Both specifications are open to industry use and collaboration. Product conformance must be checked against the specific version, deployment scope and actual evidence. [Read the security baseline and self-assessment tools](https://github.com/teloa-ai/agent-baseline-assessor) ### Get started with Teloa 0.2.2 Teloa Community edition (0.2.2) is available through **GitHub, npm and Docker Hub** for local deployment. Start with a real project: describe the goal, select materials, create the employees you need, and let the team work toward deliverables you can review and accept. [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.2) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.2) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) [Website](https://www.teloa.ai/en/) · [Documentation](https://docs.teloa.ai/v/0.2.31/markdown/en/index.md) · [Marketplace](https://market.teloa.ai/en/) · [Full release notes](https://docs.teloa.ai/v/0.2.31/markdown/en/changelog/index.md) ### Install and upgrade - Install `@teloa/cli@0.2.2` from npm, use this version’s native GitHub Release package, or run the Docker Hub image `teloa/teloa:0.2.2`. - For an existing native installation, use the new command package to run `teloa stop`, `teloa upgrade --to 0.2.2` and `teloa status`. Keep the original `--home`, installation identity, data and workspace. Retain the Compose project and volumes when upgrading Docker, and back up before switching. - This release has been verified on native macOS ARM64 and Docker Linux ARM64; amd64 and other native systems have not been verified. - Codex delivery, saving a method and running a plan manually have real-model verification. Calendar-triggered real-model execution, Claude Agent through the owner’s Anthropic sign-in, complete OpenAI/ElevenLabs speech combinations and all optional device combinations have not been fully verified. See [installation](https://docs.teloa.ai/v/0.2.31/markdown/en/start/quickstart.md) and [backup and upgrades](https://docs.teloa.ai/v/0.2.31/markdown/en/deploy/backup.md). The release documentation snapshot is retained under `/v/0.2.2/`; the root path tracks current documentation. ## Previous stable releases ::: details View the 0.2.1 stable release record ### 0.2.1 (stable) 2026-10-18 · [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.1) · [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.1) · [Docker Hub](https://hub.docker.com/r/teloa/teloa) **Teloa 0.2.1 (Community edition)** brings persistent roles, team collaboration, multiple execution harnesses, project memory, document retrieval and versioned deliverables into one workspace. Describe your goal, follow the work, handle required approvals, then collect and revise the result. #### AI employees: lasting responsibility for recurring work - Describe the role and requirements in Markdown, carry in the current business and preserve the draft. Review the proposed responsibilities and boundaries before explicitly confirming creation. - Keep an employee's responsibilities, working methods, business scope, knowledge, skills and permitted tools together across tasks. - Open employee conversations, tasks, deliverables and experience from the employee directory. Chats, employees and groups have separate lists; compact employee overviews reveal details on demand. - Adding an employee, configuring capabilities and starting execution remain separate actions. Employees work within their actual business and tool permissions. #### AI teams: combine roles into a business workflow - Discover AI teams separately in the marketplace. Team templates can include multiple employees, skills, reference materials, optional connections and work templates, reducing repetitive setup. - Review the employees and target business, then enable the connections and work arrangements you need. A template does not connect accounts, expand permissions or start background work automatically. - Collaborate with several employees in a group, use explicit `@` mentions, and retain the assignment source and result replies. Sharing a group does not grant access to every member's business materials. - **Lasting teams and temporary execution teams are different:** employees are persistent roles; execution assistants split up the current task, can work on independent parts in parallel, and return results for consolidation. They do not automatically become employees or group members. #### Multiple execution harnesses: DSH, Codex and Claude Code | Execution option | Its role in Teloa | | --- | --- | | DSH | Runs the main conversation, tools and native runtime. Teloa maintains business attribution, employee identity, permissions, approvals and delivery. This release pins `0.2.1-rc.2`. | | Codex | Receives independent work through the native Codex provider, uses its native execution environment after the owner's confirmation, and returns code, file or tool results to the original work. | | Claude Code / Claude Agent | Receives independent work through the native Claude Code provider and Claude Agent execution path. Some product labels use “Claude Agent”. | - Specify the engine in natural language: “Use Codex to review this code,” “Have Claude Agent analyze these files,” or ask for an available engine to be selected automatically. A task-specific choice does not change permanent preferences or silently replace the chat model. - **An engine is not a model service.** Integration does not include a provider account, subscription or inference allowance. Configure the authentication and model connection required by the chosen provider. Ordinary chats, groups and employees retain their working rules. - External engines use their native permissions and require the owner's consent for each launch. Teloa's file scope and per-tool approvals are not automatically passed into the external engine; review its actual working directory and permissions before consenting. - Follow actual execution and required authorization, then receive the result in the original task. If a requested engine is unavailable, Teloa preserves the task and reports the reason instead of silently substituting another engine. #### Projects and project memory: context for subsequent work - Organize project goals, briefs, tasks, employees, groups, materials, automations and deliverables, with dedicated project conversations. - Include the project goal, active rules, relevant materials and memories in subsequent model requests. Inspect what was used and its sources, beyond a list of linked objects. - Review memory candidates and sources, withdraw unsuitable content, start a conversation without long-term memory, and explicitly choose whether personal preferences apply. - Share a fixed memory revision and withdraw subsequent use. Withdrawal cannot recall information already sent to an external model. #### Document indexes and local models: retrieval and generation configured separately - Select local Markdown files explicitly, track versions, add them to an index and inspect citations and excerpt locations. Teloa does not scan the entire computer automatically. - Prepare the built-in Qwen3 Embedding 0.6B or EmbeddingGemma 2 in the Knowledge retrieval-model panel, with download confirmation and visible progress. This list is separate from the marketplace model catalog. - Use Ollama to inspect the local chat runtime, connect an existing model, confirm downloads or cancel preparation. Chat models generate answers; embedding models support retrieval. They are managed separately. - Indexing runs on the deployment device or container. Retrieved excerpts may still reach a remote chat provider. File preview supports more formats than the current local index. #### Chats and groups: distinguish stream replies, topics and assignments - Keep ordinary chats, employee conversations and groups in separate lists, preserving drafts and working context. - Replies to a group-stream message stay in the stream with their source; topic replies stay inside the topic rather than turning every response into a topic. - Explicit `@` mentions carry through to assigned work. Execution still checks membership, business scope, materials and tool permissions; a group message does not replace the owner's approval. #### Natural-language tasks and versioned deliverables - Describe work in Markdown in the main task entry points, carrying the source conversation, project or business into the draft. Saving is separate from execution. - Follow execution, handle approvals, stop work, inspect files and obtain deliverables in one working path. The companion and general Agent can work without any created employees. - Preview and download results, revise a selection or the whole deliverable from feedback, and retain v1, v2 and later versions with their original attribution. Business acceptance continues to reference its selected version. - Reuse an accepted method as a skill or plan draft. A new plan stays paused until explicitly enabled. Employee creation uses a Markdown draft; some group configuration remains structured. #### Business records, dashboards and apps - Describe a business configuration or change in natural language, review and adopt its draft, import records, connect approved sources and progress to an accepted result. - Records and Dashboards show their own pages and remember their selections. Cards remove repeated headings and explanations, fit their content, and show chart labels and values; missing data remains distinct from zero. - Discover employees and teams in the marketplace; manage installed capabilities and connections in Apps. Model setup, application setup and business permissions remain explicit rather than changing automatically when a template is added. #### Voice input, TTS and calls | Capability | What this release provides | | --- | --- | | Voice input | Local SenseVoice transcription after model preparation and microphone permission. Recognized text goes into a draft for review and sending. | | Call recognition (ASR) | MiniMax cloud recognition or local SenseVoice, configured separately from speech playback. | | Speech playback (TTS) | MiniMax, OpenAI and ElevenLabs, each with its own API key, model and voice settings, rather than one hardcoded provider. | | Voice selection and preview | Compact Mandarin/English voice lists and preview for MiniMax. OpenAI and ElevenLabs use model and voice IDs with a saved-configuration test. | | Continuous call speech | Improve transitions between consecutive short sentences. MiniMax's persistent bidirectional TTS supplies continuous audio; HTTP fallback remains sentence-based, with stop, cancellation and interruption controls. | - Call an employee or the companion, mute, interrupt a response or hang up, and keep call controls accessible during navigation. Closing details does not hang up; hanging up does not stop work already running in the conversation. - Record a companion voice message and review its transcription in the original conversation draft. Hiding the companion does not end the conversation. - Voice credentials are separate from chat-model credentials. #### Auto Dream, employee experience and a personal twin - Auto Dream organizes private daily summaries from work evidence and saves employee experience with sources and scope, with inspection and withdrawal controls. - Employee experience, project memory and personal twin materials remain distinct. Judgment samples and habit observations are private materials, not automatic context for every ordinary conversation or model-weight training. #### Files, tools and upgrade continuity - Preview Markdown, code, images, PDF, HTML, Office files and tables, inspect current-turn files and available text changes, and save a specific deliverable version when needed. - Use Workflow steps, PTC tool composition and explicitly requested Ralph iteration within their approval boundaries. Temporary delegation retains the current work scope and allowance. - Configure and authorize optional Browser Use, Computer Use and IM channels separately. Their presence in the package does not grant device access. - Preserve the original installation identity, data and workspace during upgrades, with a backup before switching. More consistent drawers, navigation, pending/failure states and reduced repeated copy retain history and unsubmitted drafts. #### Experience and reliability fixes - **Create employees in Markdown:** the main creation entry accepts one description, carries in the business and preserves drafts; writes still require preview and confirmation. - **Clearer counts:** marketplace “Added” consistently means items added on this machine; the workbench labels task counts explicitly, and deliverable counts show the actual values. - **Import and calls:** Traditional Chinese import copy respects regional terminology; narrow-window call controls avoid both quick chat and the companion. - **Recover existing work:** older assignments can recover their exact original owner message; submission failures retain safe diagnostics without automatic resending. - **Avoid blocked concurrency:** verified finished work releases its active slot while unknown token usage remains reserved. A timeout alone never releases the slot. - **Catalog and documentation:** team requirements describe Teloa permissions; release notes explain employees, teams, execution harnesses and speech services separately. #### Install and upgrade - Install `@teloa/cli@0.2.1` from npm, use this version’s native GitHub Release package, or run the Docker Hub image `teloa/teloa:0.2.1`. - For an existing native installation, use the new command package to run `teloa stop`, `teloa upgrade --to 0.2.1` and `teloa status`. Keep the original `--home`, installation identity, data and workspace. Retain the Compose project and volumes when upgrading Docker, and back up before switching. - This release has been verified on native macOS ARM64 and Docker Linux ARM64; amd64 and other native systems have not been verified. - Codex delivery, saving a method and running a plan manually have real-model verification. Calendar-triggered real-model execution, Claude Agent through the owner’s Anthropic sign-in, complete OpenAI/ElevenLabs speech combinations and all optional device combinations have not been fully verified. See [installation](https://docs.teloa.ai/v/0.2.31/markdown/en/start/quickstart.md) and [backup and upgrades](https://docs.teloa.ai/v/0.2.31/markdown/en/deploy/backup.md). The release documentation snapshot is retained under `/v/0.2.1/`; the root path tracks current documentation. ::: ::: details View the 0.2.0 stable release record ### 0.2.0 (stable) 2026-10-18 · Available on [GitHub Releases](https://github.com/teloa-ai/teloa/releases/tag/v0.2.0), [npm](https://www.npmjs.com/package/@teloa/cli/v/0.2.0) and [Docker Hub](https://hub.docker.com/r/teloa/teloa). #### AI employees: lasting responsibility for recurring work - Keep an employee's responsibilities, working methods, business scope, knowledge, skills and permitted tools together across tasks. - Open employee conversations, tasks, deliverables and experience from the employee directory. Chats, employees and groups have separate lists; compact employee overviews reveal details on demand. - Adding an employee, configuring capabilities and starting execution remain separate actions. Employees work within their actual business and tool permissions. #### AI teams: combine roles into a business workflow - Discover AI teams separately in the marketplace. Team templates can include multiple employees, skills, reference materials, optional connections and work templates, reducing repetitive setup. - Review the employees and target business, then enable the connections and work arrangements you need. A template does not connect accounts, expand permissions or start background work automatically. - Collaborate with several employees in a group, use explicit `@` mentions, and retain the assignment source and result replies. Sharing a group does not grant access to every member's business materials. - **Lasting teams and temporary execution teams are different:** employees are persistent roles; execution assistants split up the current task, can work on independent parts in parallel, and return results for consolidation. They do not automatically become employees or group members. #### Multiple execution harnesses: DSH, Codex and Claude Code | Execution option | Its role in Teloa | | --- | --- | | DSH | Runs the main conversation, tools and native runtime. Teloa maintains business attribution, employee identity, permissions, approvals and delivery. This release pins `0.2.1-rc.2`. | | Codex | Receives independent work through the native Codex provider, uses its native execution environment after the owner's confirmation, and returns code, file or tool results to the original work. | | Claude Code / Claude Agent | Receives independent work through the native Claude Code provider and Claude Agent execution path. Some product labels use “Claude Agent”. | - Specify the engine in natural language: “Use Codex to review this code,” “Have Claude Agent analyze these files,” or ask for an available engine to be selected automatically. A task-specific choice does not change permanent preferences or silently replace the chat model. - **An engine is not a model service.** Integration does not include a provider account, subscription or inference allowance. Configure the authentication and model connection required by the chosen provider. Ordinary chats, groups and employees retain their working rules. - External engines use their native permissions and require the owner's consent for each launch. Teloa's file scope and per-tool approvals are not automatically passed into the external engine; review its actual working directory and permissions before consenting. - Follow actual execution and required authorization, then receive the result in the original task. If a requested engine is unavailable, Teloa preserves the task and reports the reason instead of silently substituting another engine. #### Projects and project memory: context for subsequent work - Organize project goals, briefs, tasks, employees, groups, materials, automations and deliverables, with dedicated project conversations. - Include the project goal, active rules, relevant materials and memories in subsequent model requests. Inspect what was used and its sources, beyond a list of linked objects. - Review memory candidates and sources, withdraw unsuitable content, start a conversation without long-term memory, and explicitly choose whether personal preferences apply. - Share a fixed memory revision and withdraw subsequent use. Withdrawal cannot recall information already sent to an external model. #### Document indexes and local models: retrieval and generation configured separately - Select local Markdown files explicitly, track versions, add them to an index and inspect citations and excerpt locations. Teloa does not scan the entire computer automatically. - Prepare the built-in Qwen3 Embedding 0.6B or EmbeddingGemma 2 in the Knowledge retrieval-model panel, with download confirmation and visible progress. This list is separate from the marketplace model catalog. - Use Ollama to inspect the local chat runtime, connect an existing model, confirm downloads or cancel preparation. Chat models generate answers; embedding models support retrieval. They are managed separately. - Indexing runs on the deployment device or container. Retrieved excerpts may still reach a remote chat provider. File preview supports more formats than the current local index. #### Chats and groups: distinguish stream replies, topics and assignments - Keep ordinary chats, employee conversations and groups in separate lists, preserving drafts and working context. - Replies to a group-stream message stay in the stream with their source; topic replies stay inside the topic rather than turning every response into a topic. - Explicit `@` mentions carry through to assigned work. Execution still checks membership, business scope, materials and tool permissions; a group message does not replace the owner's approval. #### Natural-language tasks and versioned deliverables - Describe work in Markdown in the main task entry points, carrying the source conversation, project or business into the draft. Saving is separate from execution. - Follow execution, handle approvals, stop work, inspect files and obtain deliverables in one working path. The companion and general Agent can work without any created employees. - Preview and download results, revise a selection or the whole deliverable from feedback, and retain v1, v2 and later versions with their original attribution. Business acceptance continues to reference its selected version. - Reuse an accepted method as a skill or plan draft. A new plan stays paused until explicitly enabled. Some employee and group creation flows in 0.2.0 still use structured configuration. #### Business records, dashboards and apps - Describe a business configuration or change in natural language, review and adopt its draft, import records, connect approved sources and progress to an accepted result. - Records and Dashboards show their own pages and remember their selections. Cards remove repeated headings and explanations, fit their content, and show chart labels and values; missing data remains distinct from zero. - Discover employees and teams in the marketplace; manage installed capabilities and connections in Apps. Model setup, application setup and business permissions remain explicit rather than changing automatically when a template is added. #### Voice input, TTS and calls | Capability | What this release provides | | --- | --- | | Voice input | Local SenseVoice transcription after model preparation and microphone permission. Recognized text goes into a draft for review and sending. | | Call recognition (ASR) | MiniMax cloud recognition or local SenseVoice, configured separately from speech playback. | | Speech playback (TTS) | MiniMax, OpenAI and ElevenLabs, each with its own API key, model and voice settings, rather than one hardcoded provider. | | Voice selection and preview | Compact Mandarin/English voice lists and preview for MiniMax. OpenAI and ElevenLabs use model and voice IDs with a saved-configuration test. | | Continuous call speech | Improve transitions between consecutive short sentences. MiniMax's persistent bidirectional TTS supplies continuous audio; HTTP fallback remains sentence-based, with stop, cancellation and interruption controls. | - Call an employee or the companion, mute, interrupt a response or hang up, and keep call controls accessible during navigation. Closing details does not hang up; hanging up does not stop work already running in the conversation. - Record a companion voice message and review its transcription in the original conversation draft. Hiding the companion does not end the conversation. - Voice credentials are separate from chat-model credentials. #### Auto Dream, employee experience and a personal twin - Auto Dream organizes private daily summaries from work evidence and saves employee experience with sources and scope, with inspection and withdrawal controls. - Employee experience, project memory and personal twin materials remain distinct. Judgment samples and habit observations are private materials, not automatic context for every ordinary conversation or model-weight training. #### Files, tools and upgrade continuity - Preview Markdown, code, images, PDF, HTML, Office files and tables, inspect current-turn files and available text changes, and save a specific deliverable version when needed. - Use Workflow steps, PTC tool composition and explicitly requested Ralph iteration within their approval boundaries. Temporary delegation retains the current work scope and allowance. - Configure and authorize optional Browser Use, Computer Use and IM channels separately. Their presence in the package does not grant device access. - Preserve the original installation identity, data and workspace during upgrades, with a backup before switching. More consistent drawers, navigation, pending/failure states and reduced repeated copy retain history and unsubmitted drafts. #### Installation, upgrades and support - Use the stable command package to run `teloa stop`, `teloa upgrade --to 0.2.0` and `teloa status`. Keep the original `--home` for a custom installation; the upgrade retains identity and data and creates a backup before switching. Keep an existing Docker deployment's Compose project and data volumes, and back it up before updating the image. - Native macOS ARM64 and Docker Linux ARM64 are verified separately; amd64 and other native operating systems are outside this release's acceptance. Configure and check each model, speech provider and optional device combination according to its guide. A stable version number does not grant permissions or guarantee result quality. - See [installation](https://docs.teloa.ai/v/0.2.31/markdown/en/start/quickstart.md) and [backup and upgrades](https://docs.teloa.ai/v/0.2.31/markdown/en/deploy/backup.md). Stable documentation is preserved under `/v/0.2.0/`; the root path continues to track current documentation. ::: ## Historical prereleases ::: details View Alpha/Beta release history ### 0.2.0-beta.7 (preview · GitHub Release / Docker Hub) 2026-10-18 · The native package is published on [GitHub Releases](https://github.com/teloa-ai/teloa/releases/tag/v0.2.0-beta.7) and has been verified on macOS arm64 with Node.js 24. The Docker image has passed an isolated startup check on Linux ARM64. npm still provides 0.2.0-alpha.7. It includes everything in 0.2.0-beta.6. This historical version collected Beta-stage business, marketplace and installation fixes. The complete capability overview for stable `0.2.0` is in the section above. #### Latest fixes - Fix: business pages now separate Records from Dashboards, and each tab remembers its own page. Dashboard cards show their title once, state shared notes once per page (cards still note their own missing or truncated data), size to their content, and label chart bars with categories and values. Pages created from an official template show their built-in page names in the interface language, while names you changed stay as you set them. - Fix: in the marketplace, a team's skills and connections are counted separately, a team added to this computer shows Added with Continue adding, counts use the singular form for one item, buttons no longer wrap their labels in English, and the template import dialog wording was corrected. - Upgrades: run `teloa upgrade --to 0.2.0-beta.7` from the new command package to upgrade an existing installation, such as `0.2.0-beta.6`, `0.2.0-beta.5` or `0.2.0-alpha.7`, while keeping its identity and data. Known limitations are the same as in 0.2.0-beta.4. ### 0.2.0-beta.6 (preview · superseded by 0.2.0-beta.7) 2026-10-18 · The native package is published on [GitHub Releases](https://github.com/teloa-ai/teloa/releases/tag/v0.2.0-beta.6) and has been verified on macOS arm64 with Node.js 24. It is not published to npm or Docker Hub. It includes everything in 0.2.0-beta.5. - Fix: in-app guidance now points to the places that exist today: resume a newly added employee from Employees, open files and skills from Apps, reopen a configuration draft from Added, and the Auto Dream notice opens the employee it refers to. Wording that still said solution now says template, and several English labels in the work results panel and import forms were corrected. - Upgrades: run `teloa upgrade --to 0.2.0-beta.6` from the new command package to upgrade an existing installation, such as `0.2.0-beta.5`, `0.2.0-beta.4` or `0.2.0-alpha.7`, while keeping its identity and data. Known limitations are the same as in 0.2.0-beta.4. ### 0.2.0-beta.5 (preview · superseded by 0.2.0-beta.6) 2026-10-18 · The native package is published on [GitHub Releases](https://github.com/teloa-ai/teloa/releases/tag/v0.2.0-beta.5) and has been verified on macOS arm64 with Node.js 24. It is not published to npm or Docker Hub. It includes everything in 0.2.0-beta.4. - Fix: when another add of the same catalog employee is still installing its methods, adding that employee no longer returns early as pending; it continues to the same method installation and returns the complete receipt for the same employee. - Upgrades: run `teloa upgrade --to 0.2.0-beta.5` from the new command package to upgrade an existing installation, such as `0.2.0-beta.4`, `0.2.0-beta.3` or `0.2.0-alpha.7`, while keeping its identity and data. Known limitations are the same as in 0.2.0-beta.4. ### 0.2.0-beta.4 (preview · superseded by 0.2.0-beta.5) 2026-10-18 · The native package is published on [GitHub Releases](https://github.com/teloa-ai/teloa/releases/tag/v0.2.0-beta.4) and has been verified on macOS arm64 with Node.js 24. It is not published to npm or Docker Hub. It includes everything in 0.2.0-beta.3. - Fix: running a plan right away while the scheduler is recovering the same run no longer reports a conflict, marks the plan as failing or leaves an extra session; the run starts once. - Upgrades: run `teloa upgrade --to 0.2.0-beta.4` from the new command package to upgrade an existing installation, such as `0.2.0-beta.3` or `0.2.0-alpha.7`, while keeping its identity and data. Known limitations: saving a confirmed deliverable's method as a plan from the conversation, running it once manually, opening the result and completing the task have been verified with a real model; a run started by the calendar schedule itself has not. Other limitations are the same as in 0.2.0-beta.3. ### 0.2.0-beta.3 (preview · superseded by 0.2.0-beta.4) 2026-10-18 · The native package is published on [GitHub Releases](https://github.com/teloa-ai/teloa/releases/tag/v0.2.0-beta.3) and has been verified on macOS arm64 with Node.js 24. It is not published to npm or Docker Hub. See [Install and run Teloa](https://docs.teloa.ai/v/0.2.31/markdown/en/start/quickstart.md) for installation and upgrades. - Runtime: DSH is pinned to `0.2.1-rc.2`. - Upgrades: run `teloa upgrade --to 0.2.0-beta.3` from the new command package to upgrade an existing installation, such as `0.2.0-alpha.7`, while keeping its identity and data. Other versions still upgrade from the npm registry. - Work and deliverables: assign work to the general agent or an AI employee and follow it in the original conversation. Deliverables keep versions; give feedback or revise a selected part, and earlier versions stay readable. - Reusable methods: save a confirmed way of working as a skill, or as a plan that stays paused until you enable it. - Collaboration: group conversations carry explicit `@` mentions through the work; employee chats are kept separate from general conversations. - External coding agents: run Codex or Claude Agent tasks from the workbench with native approvals, using credentials you configure. - Web search and fetch: optional DeepSeek search and page fetch in Settings > Connections and runtime > Web search, with your own credentials. - Market: one flow to create employees and teams; installing an employee brings its complete resources; reviewed app packs; share an AI employee as a template. - Business: local business work with approved record changes, and dashboards that reflect task results. - Voice: local speech recognition, and speech playback with MiniMax, OpenAI or ElevenLabs, including compact bilingual voice selection, preview and continuous playback. Known limitations: saving a confirmed deliverable's method as a plan through the conversation, enabling it and its first scheduled run have not been verified end-to-end with a real model. In complex analyses, model conclusions can claim more than the evidence supports; review deliverables before relying on them. Claude Agent with Anthropic models through your own account sign-in has not been verified. ### 0.2.0-beta.2 (preview · superseded by 0.2.0-beta.3) 2026-10-18 · The first public 0.2.0 Beta preview. Its [GitHub Release](https://github.com/teloa-ai/teloa/releases/tag/v0.2.0-beta.2) is unchanged. Its command package cannot upgrade an existing installation to its own version; use `0.2.0-beta.3` instead. ### 0.2.0-alpha.7 (published · Alpha) 2026-10-01 · The exact version is published on [npm as `@teloa/cli`](https://www.npmjs.com/package/@teloa/cli) and [Docker Hub as `teloa/teloa`](https://hub.docker.com/r/teloa/teloa). Docker has been verified on Linux ARM64; amd64 has not been verified. - Build and verify an SOC business from the business page: connect a read-only alert source, declare its mapping and enable sync, assign a real Task/Run from a record, then see its progress and the exact version of the accepted deliverable on the same business page. Other industries are examples without end-to-end acceptance. - CSV and XLSX imports preview data, write a fixed batch, and recover the same request after a lost response or restart. Sync rules are controlled per mapping; equal external IDs from different sources do not overwrite local records. - Includes projects, task teams, updated extension management and composer improvements, along with marketplace, connection, local-model and extension work. External connections and extensions still need configuration and authorization. :::