Contextlattice

Name: Contextlattice
Author: sheawinkler

@sheawinkler

Repository Homepage

About Contextlattice

ContextLattice is the local-first control plane for long-horizon agent memory, coordination, and behavioral provenance.

Basic information

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "contextlattice": {
      "type": "streamable-http",
      "url": "http://127.0.0.1:8075/mcp",
      "headers": {
        "Accept": "application/json, text/event-stream",
        "Content-Type": "application/json",
        "x-api-key": "${CONTEXTLATTICE_ORCHESTRATOR_API_KEY}"
      }
    }
  }
}

Tools

No tools detected

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is ContextLattice?

ContextLattice is a private-by-default memory and context orchestration system for AI agents. It provides a unified write/read contract for agentic memory, durable fanout across retrieval and storage lanes, and staged retrieval with agent sessions, all running in a local-first deployment with optional hosted surfaces. It is implemented as an MCP HTTP Gateway with a Go/Rust runtime.

How to use ContextLattice?

Clone the repository, copy .env.example to .env, then run gmake quickstart to launch the selected stack. Verify with curl -fsS http://127.0.0.1:8075/health | jq and the proof scripts. AI agents can install and integrate by following the Agent Quickstart block, which uses contextlattice_adopt integrate to write managed instruction blocks for supported agent profiles.

Key features of ContextLattice

Unified write/read memory contract for agentic systems.
Durable fanout across retrieval and storage lanes.
Staged retrieval with fast now and deep continuation.
Agent sessions with prior work, objective lineage, and handoffs.
Go/Rust runtime ownership for the active application path.
Local-first deployment with optional hosted surfaces.

Use cases of ContextLattice

Integrate memory and context orchestration into existing agent workflows (Codex, Claude Code, OMP, Mercury, etc.).
Manage long-running agent sessions with checkpoints, handoffs, and run-card exports.
Run a private-by-default memory layer for agentic systems without third-party cloud dependencies.
Deploy a single-container lite stack for local agent memory with topic_rollups and Qdrant.

FAQ from ContextLattice

What runtime dependencies does ContextLattice require?

The public lite stack uses Go and Rust services; the legacy Python runtime is archived and optional. Single-container lite builds run gateway-go without Python. Qdrant is the default vector lane; pgvector is supported but not started by default in lite mode.

How do I install and launch ContextLattice?

Clone the repository, copy .env.example to .env, then run gmake quickstart. The quickstart prompts for the runtime profile and launches the selected container stack. Verify with curl -fsS http://127.0.0.1:8075/health | jq.

What license is ContextLattice released under?

ContextLattice is licensed under BSL 1.1.

What is the default embedding path?

Embedding defaults to the Rust fastembed-rs sidecar. Ollama is available as an explicit compatibility fallback but is not the preferred embedding path.

How does the model inference runtime work?

The runtime detects the host profile and probes local backends. Default priorities vary by platform (Apple Silicon: mlx, vllm-metal; CUDA/ROCm: sglang, vllm; generic CPU: openai-compatible, llama.cpp). The provider is controlled by ORCH_INFER_PROVIDER=auto and can be overridden with environment variables.