Trilith v0.1 beta — context engineering for agents (budgeted memory, 3 tiers, audit trail). Core is Python; use via REST or adapters (LangGraph / OpenAI Agents / Claude / TS). Needs trilith serve loca
Budgeted context. Three tiers. Auditable assembly. Open-source context management layer for AI agents — sits between your agent and the LLM prompt.
Agents drown in context. Trilith governs what enters the prompt:
Tier Role ------------ Semantic Durable facts Procedural Task steps (fold into summaries) Episodic Scoped events + cascading forget
Policy → rank → fill budget → return included + excluded with reasons. No silent drops.
Expect - Local SQLite + TF-IDF ranking (zero external AI dependency) - REST (:8080) + gRPC (:50051) + MCP - Plug-and-play adapters: LangChain/LangGraph, OpenAI Agents, Claude tools, TypeScript SDK - Auditable excludeditems with reasons - Physical forget(scope) across tiers
Don’t expect (yet) - Production multi-tenant IAM / real tenant IDs (scopes are coarse enums) - Vector DB backends - Auth on REST/gRPC — bind to 127.0.0.1 only for beta; do not expose ports publicly - PyPI / npm packages yet — install from this GitHub repo (see below)
- No API keys required for Trilith itself (it does not call an LLM) - No authentication on REST/gRPC yet → local testing only - Prefer: trilith serve --host 127.0.0.1 ...
There are several strong memory solutions for AI agents already. Here's an honest comparison — including where they beat Trilith.
Trilith Mem0 Zep LangMem --------------- Architecture 3-tier (semantic / procedural / episodic) with an explicit context budget LLM-driven classification, hybrid vector + graph Temporal knowledge graph LangGraph-native store, 3 memory types Framework lock-in None — LangChain, LangGraph, OpenAI Agents SDK, Claude to
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/Umairkhan2324--trilith.html)
Open-source control plane for Codex projects: Git-backed context, visible agent progress, scoped MCP access, resumable work, and safe handoffs.
Koragraph/KoragraphMCPThe memory layer that gets better as you use it. What broke, why, what depended on what, even the "push this to Wednesday" kind of thing that never makes it into any doc. No one has to write it down.
Ignitic-AI/igniticAI agents that run your e-commerce ops — Shopify, ads, CRM & support via 190+ MCP tools. Open-source, self-hosted, one docker compose up.
mrcbrbn5361/SyncytiumMDUniversal Context, Rules & Handoff Bridge for AI Coding Tools (Cursor, Copilot, Cline, Claude Code, Antigravity, MCP)
CTRLRun/ctrlrunThe execution safety layer for AI agents.
Novaya-AI/novgraphOpen-source client for Novgraph, Novaya's proprietary hosted code knowledge graph. Gives coding agents recorded intent, related-code context, ranked change impact and architecture context over MCP or
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.