36_chambers
Federated knowledge retrieval engine for agents that uses Jev to route queries to backends and re-rank evidence, with keyword fallback.
Overview
36 Chambers is a registry-first retrieval engine for autonomous agents. Data stays in its native stores (SQLite FTS5, ChromaDB, a knowledge graph) and agents query them through one immutable registry. It is read-only by default and aimed at people building agent memory or research tools who want grounded, cited answers.
Since v0.2.0, Jev is an optional router and re-ranker. It decomposes a query into chamber routes with confidence scores; routing goes through Jev when confidence is at least 0.6 and falls back to keyword matching on low confidence or API errors. Retrieved evidence is re-ranked with Jev Noul judgments, an HTTP Cohere-compatible reranker or a local BGE cross-encoder. An opt-in extractive synthesis step assembles answers only from retrieved sentences and flags missing evidence when sufficiency is below 0.5.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which chamber (backend) should handle this query? | FTS5 · Chroma · Graph · ... |
| Yes / no | Is this evidence relevant to the query? | yes · no |
How it works
- Receive a query from CLI, API or agent
- Jev routes it to chambers; below 0.6 confidence, use keyword routing
- Query the federated adapters
- Re-rank evidence by Jev Noul probability (or HTTP/BGE fallback)
- Optionally synthesize a cited extractive answer, gated on sufficiency
Strengths
- Fail-soft: works without a TypeSafe key using deterministic routing
- Confidence threshold decides when to trust the model
- Local vectors with telemetry disabled
Caveats
- PolyForm Noncommercial licence; commercial use needs an agreement
- Zero GitHub stars at listing time; early project
Getting started
pip install -r requirements.txt, then set TYPESAFE_API_KEY to enable Jev routing (optional).At a glance
- Type
- Library
- Status
- prototype
- Language
- Python
- Runtime
- CLI / library
- Has code
- yes
- License
- PolyForm-Noncommercial-1.0.0
- Stars
- 0
Use cases and integrations
How curated lists describe it
Federated knowledge retrieval engine for AI agents. Routes queries across specialized backends (FTS5, ChromaDB, graph) via a registry-first orchestrator with TypeSafe/Jev AI routing and evidence re-ranking. Read-only by default, security-gated by design.
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide here?
Which backend chambers a query goes to, and how relevant each piece of evidence is for re-ranking.
Does it need a TypeSafe key?
No; without one it uses deterministic keyword routing and non-Jev rerankers.
Can I use it commercially?
Not under the default PolyForm Noncommercial licence; a separate agreement is required.
Related projects
Android AI chat client and SillyTavern-compatible endpoint where Jev screens memories for relevance and serves as a judge tool.
LibraryproductionKotlinLocal long-term memory for coding agents with an optional Jev reranker that orders recalled memories.
ProjectproductionTypeScriptVolcengine's context database for agents, with a Jev rerank client that scores candidate documents with jev-latest as relevance.
LibraryproductionPythonAgent harness for long-running jobs that survive across sessions; listed for a Jev-backed table that routes requests to models.
ProjectbetaTypeScriptSkill pack handing an agent's small decisions to Jev: model routing, skill choice, search, memory, compaction and computer use.
Agent skillbetaPythonFilters an assistant's memories by relevance, not resemblance: one Jev yes/no per memory in a single request, kept above a threshold.
ProjectbetaPython