Build, inspect, and serve context deterministically for AI agents and LLMs.
Context Engine provides reproducible, auditable, and offline-first context delivery.
It ensures that the same query over the same cache produces identical outputs across:
- hardware architectures (x86_64, aarch64)
- operating systems (Linux, macOS, Windows)
- supported Rust compiler versions
No randomness. No hidden state. Fully explainable.
The deterministic engine powering the platform.
Content-addressed documents, immutable caches, and reproducible selection logic for LLMs.
Command-line interface to build, inspect, and verify context caches.
Supports CI/CD pipelines and local audits of agent behavior.
MCP server exposing context caches to agents via JSON-RPC 2.0 over stdio.
Deterministic responses ensure agents always get the same context.
Compatibility harness validating determinism, schema compliance, and backward compatibility.
Runs externally via CLI and MCP; no internal crate dependencies.
Formal specifications and invariants for the platform.
Includes frozen JSON schemas, MCP error contracts, and selection behavior rules.
Documentation, how-to guides, and usage examples for developers and integrators.
JavaScript SDK (work in progress).
Provides a programmatic interface to build and resolve context in Node.js.
Python SDK (work in progress).
Programmatic access for Python-based AI workflows.
- Deterministic Selection: Same inputs → identical outputs.
- Content-Addressed: SHA-256 hashed documents ensure integrity.
- Token-Budget First: Designed for LLM window constraints.
- Offline & Auditable: No network dependencies; inspectable caches.
- Frozen Contracts: Outputs and errors are schema-locked and versioned.
- Enterprise AI deployments
- CI/CD pipelines for LLM-based systems
- Air-gapped and on-prem AI infrastructure
- Auditable, regulatory-compliant AI workflows
- Multi-agent systems requiring reproducible behavior
All open-source repositories are licensed under Apache License 2.0.
The "Context Engine" trademark is reserved for the contributors.
Learn more: Visit the Context Engine docs for guides, specs, and examples.