crewcontext 0.1.0

Auditable shared memory for AI agent systems

CrewContext

Auditable shared memory for AI agent systems.

The Problem

Multi-agent AI systems break at handoffs. Agent 1 processes an invoice. Agent 2 validates it. Agent 3 reconciles discrepancies. But Agent 3 has no idea what Agent 1 found. Context is lost. Decisions are invisible. Nothing is auditable.

Existing agent frameworks give you orchestration — but not memory. Chat history is personal. RAG is read-only. Neither gives you a shared, structured, temporal record of what happened, who did it, and why.

In regulated industries — finance, insurance, compliance — this isn't just inconvenient. It's a liability.

What CrewContext Does

CrewContext is a context coordination layer that sits underneath your agent framework and provides:

• Shared event store — Every agent action is recorded. Nothing is lost at handoffs.
• Causal DAG — Every event tracks what caused it. You can answer "why did this happen?" by walking the chain backwards.
• Temporal queries — Reconstruct the exact state of any entity at any point in time. "What did we know at 2pm yesterday?"
• Versioned entities — Business objects (invoices, customers, claims) are snapshotted at each stage, never overwritten.
• Policy router — Deterministic, auditable routing rules with composable conditions. No black boxes.
• Provenance tracking — Every event records which agent, what scope, and when. Built for auditors.

Architecture

                    ┌──────────────────────────────────┐
                    │        ProcessContext API        │
                    │  emit · query · timeline · causal│
                    └──────────┬───────────────────────┘
                               │
              ┌────────────────┼────────────────┐
              │                │                │
    ┌─────────▼──────┐   ┌─────▼──────┐  ┌──────▼──────┐
    │  PostgreSQL    │   │   Neo4j    │  │   Policy    │
    │  Event Store   │   │   Graph    │  │   Router    │
    │                │   │            │  │             │
    │  Append-only   │   │  Lineage   │  │  Rules      │
    │  Temporal      │   │  Causal    │  │  Pub/Sub    │
    │  Causal links  │   │  DAG       │  │  Routing    │
    │  Versioned     │   │  Typed     │  │  decisions  │
    │  entities      │   │  relations │  │             │
    └────────────────┘   └────────────┘  └─────────────┘
         (truth)          (optional)      (in-process)

PostgreSQL is the source of truth — append-only event log, versioned entity snapshots, causal link table. Neo4j is an optional projection for graph queries and lineage visualization. Policy Router evaluates events against composable rules in-process.

Who It's For

CrewContext is for teams building multi-agent systems where trust, auditability, and context preservation matter:

• Financial operations — Payment processing, reconciliation, dispute resolution
• KYC/AML compliance — Auditable decision trails for regulators
• Insurance claims — Multi-stage pipelines where context loss means money lost
• Supply chain — Order-to-delivery orchestration across multiple agents
• Any regulated workflow where "the AI decided" isn't a good enough answer

Framework-Agnostic

CrewContext is not a replacement for your agent framework. It's the memory layer underneath it. It works with:

• CrewAI
• LangGraph
• AutoGen
• OpenAI Agents SDK
• Custom agent systems

Getting Started

pip install crewcontext
docker compose up -d
crewcontext init-db
crewcontext demo vendor-discrepancy

Full API documentation and examples are available in the docs directory.

Configuration

Environment Variable	Default	Description
CREWCONTEXT_DB_URL	postgresql://crew:crew@localhost:5432/crewcontext	PostgreSQL connection
CREWCONTEXT_NEO4J_URI	bolt://localhost:7687	Neo4j Bolt endpoint
CREWCONTEXT_NEO4J_USER	neo4j	Neo4j username
CREWCONTEXT_NEO4J_PASSWORD	crewcontext123	Neo4j password

Neo4j is optional. Pass enable_neo4j=False for Postgres-only mode.

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT