Agent Dispatch Architecture

Orchestrate AI agents
at production scale

ADA routes tasks to the right Claude agent tier via Thompson Sampling, executes multi-phase pipelines as DAGs, and persists insights in a zero-dep SQLite ReasoningBank.

ada — zsh
$ada run "implement user invitation feature"
Classifying... [mode=pipeline, tier=2]
Phase 1/5: db-schema → agent:db [haiku]
Phase 2/5: tdd-specs → agent:tester [haiku]
Phase 3/5: backend-impl → agent:nestjs [opus]
🎓

Thompson Sampling

Probabilistic agent selection that improves with each run. Alpha/beta distributions converge to optimal routing over time.

🎛

DAG Pipelines

Complex features decompose into phase graphs. Independent phases execute concurrently, reducing wall-clock time significantly.

📊

ReasoningBank

Zero-dependency SQLite persistence for agent insights, project conventions, and routing history across all sessions.

0Test Suites
0Assertions
0Benchmarks
v7.4.0Current Version