Packages
API reference for every consensus-tools package.
The consensus-tools monorepo is organized into five tiers (0 through 4). Dependencies flow strictly downward: a package in Tier 2 may depend on Tier 0 and Tier 1, but never on Tier 3 or 4. This layering is enforced in CI via dependency-cruiser.
Which package do I need?
I want to...
- Wrap any AI agent with 3 lines of code → universal
- Add consensus to a function call → wrapper
- Run a full job board with agents → core + policies
- Evaluate AI agent outputs → evals
- Gate risky AI actions → guards
- Build DAG workflows → workflows
- Manage persona reputation → personas
- Choose a storage backend → storage
- Expose consensus via HTTP API → sdk-node
- Use consensus in Claude Code → mcp
- Just need the types → schemas
Package table
| Tier | Package | Description |
|---|---|---|
| 0 | schemas | Zod schemas and TypeScript types |
| 0 | secrets | AES-256-GCM credential encryption |
| 1 | guards | Guard evaluation engine |
| 1 | telemetry | OpenTelemetry tracing and metrics |
| 1 | evals | AI agent output evaluation framework |
| 1 | personas | Persona packs, reputation engine, and respawn |
| 1 | storage | IStorage interface with JSON, SQLite, and memory backends |
| 1 | integrations | GitHub and Linear adapters |
| 1 | notifications | Multi-channel HITL dispatch (Slack, Teams, Discord, Telegram) |
| 1 | sdk-client | Lightweight client SDK |
| 2 | core | Job engine and consensus ledger |
| 2 | policies | 9 pluggable consensus resolution algorithms |
| 3 | workflows | DAG workflow engine |
| 3 | wrapper | Runtime decision firewall for function calls |
| 4 | universal | 3-line governance facade with LLM Persona mode |
| 4 | sdk-node | Node.js server SDK with HTTP API |
| 4 | mcp | Model Context Protocol server (29 tools) |
| 4 | openclaw | Open consensus protocol implementation |
| 4 | cli | Command-line interface |
For how these tiers are enforced, see Architecture.