Resolves 8 community issues + 3 PRs merged from trusted contributors. 2585 tests passing, zero clippy warnings, zero regressions.
Agents
- #890 New
agent_activatetool andPOST /api/agents/{id}/activateendpoint. Orchestrator agents can wake suspended or crashed peers. Idempotent on Running, rejects Terminated. - #868 cloneAgent API.
POST /api/agents/{template_id}/clonedeep-clones a template, applies overrides, assigns a fresh agent ID and workspace path. ExcludesMEMORY.mdso the new agent starts fresh. - #869 Server-assigned message IDs. Every
Messagenow carries a UUID v4msg_idstamped at construction, plus an optionalprovider_msg_idfor debugging. Legacy on-disk sessions get fresh IDs on first load.
Web Dashboard
- #1026 Inferencing indicator. Live pulse next to agents currently calling the LLM. Computed from
running_tasksDashMap. - #931 Providers screen overhauled. Search, filter by status, group by category (frontier / oss / aggregator / regional / local), configured providers first.
- #915 Telegram
user_idsurfaced.metadata.telegram_user_idand prompt prefix[From: Name (tg_id:NNN)]give agents a stable per-user identifier.
Channels
- #1146 Lenient binding parse. One malformed binding entry no longer drops the entire bindings table. Bad entries are logged at ERROR and dropped, survivors pass through.
Docs
- #1097 Workspaces gap documented (multi-component refactor needed, design comment posted).
- #1037 follow-up.
docs/configuration.mdauth posture wording corrected. Middleware fail-closes on authenticated routes but a small set of public routes (health, static, OAuth callback) remain reachable so the login page can render.OPENFANG_ALLOW_NO_AUTHopt-in still required for open access. - #1175 Getting-started doc links fixed.
Dependencies
- open 5.3.3 to 5.3.4
Stats
- 2585 tests passing (+42 from v0.6.4)
- Zero clippy warnings (
--workspace --all-targets) - Zero regressions
Full Changelog: v0.6.4...v0.6.5