A public, reusable starting point for autonomous agents and multi-agent systems. Generate a repository from this template, then rename it using the conventions below.

Naming

agent-<name> — autonomous agent.
swarm-<name> or orch-<name> — orchestrator or multi-agent system.
eval-<agent-name>-benchmark — dataset or quality benchmark.

Layout

Prompts are versioned separately from implementation. src/tools contains integrations, src/memory contains retrieval/state adapters, src/core contains the agent graph, and evals contains deterministic quality checks.

Quality gate

Every pull request runs .gitea/workflows/eval-quality.yml. The evaluator must keep Accuracy and Tool Calling Success at or above 95%. A failing evaluator blocks the protected main branch. The sample evaluator is offline-safe; replace its cases and assertions for a real agent.

Secrets

Never commit provider or database credentials. Configure organization-level Actions secrets in Gitea and reference them as ${{ secrets.OPENAI_API_KEY }}, ${{ secrets.TAVILY_API_KEY }}, ${{ secrets.VECTOR_DB_URL }}, and ${{ secrets.REGISTRY_TOKEN }} only when a project actually needs them.

Local development

python -m evals.run_evals --threshold 0.95
python -m src.main