Release notes — week of 11 August 2026¶
What's new this week¶
GitHub App push triggers¶
GitHub push events can now automatically dispatch a DevTeam run in your workspace. Configure a goal, model, and branch filter per installation — any matching push fires the run without any manual action.
Typical use cases:
- Automatic code review on every push to main
- Test gap analysis when a feature branch is updated
- Documentation refresh triggered by a merge
Configure in Settings → GitHub on the installation card, or via PATCH /github/installations/{id}/push-trigger. See GitHub App integration for the full reference.
Engine runs with remote repo context¶
POST /engine/run now accepts a repo_url field. When provided and source_dir is not set, the engine runner clones the repository into a temporary directory before starting the capability loop, then cleans up automatically when the run completes.
curl -X POST https://antcrew.org/engine/run \
-H "X-Api-Key: acw_..." \
-H "Content-Type: application/json" \
-d '{
"goal": "Add OpenAPI documentation to all public endpoints",
"model": "claude:claude-sonnet-5",
"repo_url": "https://github.com/my-org/my-api"
}'
The URL is validated against the platform's external URL allowlist before cloning. Clone timeout is 120 seconds; failures are logged and the run proceeds without source context rather than aborting.
Platform-wide default model (admin)¶
Platform admins can now set a platform-wide default model that applies to all workspaces with no workspace-level preference. Previously the hard-coded fallback was always claude.
Configure in Admin → Platform defaults, or:
curl -X PATCH https://antcrew.org/admin/agent-models/defaults \
-H "Content-Type: application/json" \
-d '{"default_agent_models": {"default": "groq:llama-3.1-70b-versatile"}}'
The Settings model matrix now shows a read-only plataforma row so workspace members can see the platform default before setting their own.
Precedence: per-run override → workspace agent default → workspace global default → platform default → claude.
Discover: model selector filtered to active providers¶
The model selector in the Discover page now shows only the providers your workspace has API keys for. If your workspace has groq and deepseek keys configured, only Groq and DeepSeek models appear — alongside claude (always available via the managed key) and simulated (always available for testing).
Previously the selector showed all 16 providers regardless of what was configured.
What else shipped this week¶
Billing automation example (Font Jardineria)¶
A new example in the antcrew SDK repo (examples/09_billing_automation.py) shows a realistic invoice automation workflow for a landscaping company. Demonstrates parsing free-form job notes into structured billing line-items, calculating totals with VAT, and generating an invoice. Runs with SimulatedLLM out of the box; comments show how to swap in Claude, GPT-4o, or Gemini.
Providers reference expanded¶
docs/engine/providers.md now lists all 16 supported providers with their prefix, BYOK key name, and an example model string. Previously only 6 were documented.
Upgrading¶
For platform self-hosters: pull the latest image and restart. Alembic runs migration 059 (push_goal, push_model, push_branch_filter columns on github_installation) automatically on startup.
Next release notes: week of 18 August 2026.