Docs

English

简体中文

English

简体中文

Appearance

Community open-source Web · 0.2.8 release docs · 0.2.8Content reviewed: 2026-10-23View Markdown

1.2 Get started with 0.2.8 ​

0.2.8 closes three gaps: execution controls and automatic delivery when business systems call employees, the complete employee-package material path, and day-to-day entry points for marketplace creation and Agent presets. For an existing installation, follow Upgrade and rollback to preserve its data and installation identity. You do not need to recreate employees or migrate conversations.

1. Employee API: choose the execution mode and model per request ​

Open Settings > Employee integrations, then create or edit an integration with its employee, business scope, expiry and read/execute/cancel permissions. When starting work, POST /api/v1/works and continuation requests may also include:

  • permissionMode: danger-full-access, workspace-write or read-only. New work defaults to danger-full-access when omitted; continuation inherits the previous round when omitted.
  • model: {provider, model, reasoningEffort?}. When omitted, the employee's configured default model is used.

The mode only sets the execution boundary for that round. It does not bypass the integration's execute permission, the employee's tool grants or workbench approvals. The host still validates the current preset and model catalog; an unavailable choice fails explicitly instead of silently switching.

The default is the widest execution boundary

When permissionMode is omitted for new work, it defaults to danger-full-access. Use that only when the task genuinely needs full host execution; business systems should normally pass workspace-write or read-only explicitly after confirming the employee tool grants and approval rules.

For example, to minimize write scope:

json
{"employeeId":"<employee-id>","input":"Check this order and report the result","permissionMode":"workspace-write"}

A fixed successful result now creates a formal artifact, completes the linked task and updates work progress in one transaction. Callers read the fixed text and checksums through result; a successful execution is not business acceptance.

Employee API / SDK / MCP guide

2. Employee package materials: preview, role-scoped retrieval, bulk enrollment ​

Materials in an employee package are no longer only file names:

  • Open a material from the employee details or a group source to read the body and version before deciding whether to cite it.
  • In an employee chat, retrieval follows that employee's role material scope. Other people's materials on the same machine do not widen it silently.
  • The employee group in the library now includes Add all to search (N): enroll that employee's active materials by source in one action, de-duplicated by source. If the retrieval model is not prepared, Teloa guides you to prepare it instead of silently skipping.

Employee-package update and unload keep identity, history, shared-material protection and transactional switching. The unload impact preview distinguishes what will be cleaned up from what remains.

3. Marketplace creation: skills and apps ​

The marketplace Create menu now contains:

  • Create AI employee
  • Create AI team
  • Create skill
  • Create app
  • Import template
  • Import skill

Skill and app creation still opens the editing and enablement flow in the workbench; the entry point does not install automatically, expand permissions or replace review.

4. Agent presets: standard is built in, PTC names are accurate ​

  • Teloa standard mode is now grouped under built-in presets; the existing standard mode remains so current sessions do not break.
  • PTC code modes are:
    • Code mode (retained variables): retain variables, functions and imported modules across run_code calls in the current Agent's code session; memory is not persisted and is cleared on restart, cancellation, timeout or Agent restart.
    • Code mode (Python): run run_code with Python under the current file permissions and approval rules.

A preset determines the execution method; employee tool grants determine whether a specific tool can be used. Neither replaces the other. See Native capabilities and Execution permissions.

5. Settings and sidebar ​

Settings categories now have dividers and hierarchy, making Daily preferences, Models and services, Capabilities and connections, Working style, and Advanced easier to scan. In non-Chinese locales the sidebar brand no longer overlaps the new/search actions; rely on the actual layout at narrower widths.

6. Security baseline ​

The container runtime and documented installation baseline for this release is Node.js 24.22.1. Back up before upgrading; check the native Node requirement separately from the runtime inside the Docker image, as described in Installation and running.

Next steps ​

UI details follow the installed version; the entry points in this page correspond to 0.2.8.