Deep Dives

Architecture and operational deep dives for planners, search, caching, analytics, and migration.

• Two-Model Architecture — Large-planner + small-executor split and cost model.
• Workflow Architecture — DAG execution, dispatch, retries, and safety rails.
• Registry Discovery — Tool registry shape, discovery paths, and search integration.
• Search Tuning — BM25 + embeddings hybrid, thresholds, and ranking knobs.
• Prompt Caching — Layered caching and provider behaviors (Anthropic/OpenAI).
• Analytics Guide — Metrics backends, schema, and when to pick Prometheus/OTLP/SQLite.
• Skill Library — Managing reusable skills and catalog hygiene.
• Small Model Improvements — Using lightweight models for execution and cost savings.
• Hybrid Model Routing — Route tasks across small and large models with thresholds and fallbacks.
• Control Flow Patterns — Sequence/branch/map-reduce, retries, and circuit breakers.
• Agent Delegation — Hand off tasks across specialized agents with A2A.
• Multi-Agent Coordination — Fan-out/fan-in with quotas, idempotency, and aggregation.
• Error Recovery — Classify failures, backoff retries, and compensating actions.
• Skills Packaging & Reuse — Versioned tool reuse via the skill library.
• Plugin Extension — Extend ToolWeaver at runtime with plugins.
• Idempotency & Retry — Guardrails for parallel dispatch and safe retries.
• REST API Usage — Call tools over HTTP from external clients.