User repository on GitHub
A TypeSafe-native (Jev) coding agent built on Recursive LLM Context Decomposition (RLCD), with a native macOS client
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Practices & Patterns
Jet
Coding agent that pairs explicit SQLite-typed context chunks with a small TypeSafe judgment model, so the large generation model only sees selected context.
- Stars
- 4
- Language
- Python
- Published by
- Community
- Author
- arczhi
- Last verified
- Oct 2026
About this project
A fast coding agent built on TypeSafe judgments and Recursive LLM Context Decomposition (RLCD) . Named for takeoff speed: a small judgment model decides in milliseconds, the big model only sees key context. 🧭 Open the visual design guide · 📖 How jet works — the ideas behind it — a plain-language walkthrough of the core designs (RLCD context engine, judgment-first routing, async plans, verification gates, policy and observability).
Quoted from the project's README. Read the full README.
- Languages
- Python
Highlights
- RLCD stores context as typed chunks in a SQLite tree scored by a judgment model.
- Judgment-first routing covers context selection, tool routing, gating and verification.
- Pluggable judgment and LLM providers with per-session JSONL tracing and cost accounting.
- Native macOS client with plan, context and session panes plus approval dialogs.
Watch out
Licence is CC BY-NC 4.0 so commercial use is not permitted, and first launch needs DeepSeek plus TypeSafe keys. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
A fast judgment model scores chunks as hide, note, summary or full text, and the big model only receives what matters. There is no transcript compaction step.
Generation uses DeepSeek, OpenCode Go or any OpenAI-compatible endpoint, while judgment uses TypeSafe Jev or laya-mlx. First launch prompts for the DeepSeek generation key and the TypeSafe judgment key.
Yes. Small-model answers are disk-cached by content hash, so identical decisions are not paid for twice. The digest cites 17 of 22 cache hits on a real run.
Tools are policy-gated with human approval dialogs, deterministic rules first, judgment can only tighten, and everything fails closed. Cross-model verification checks the generator with a different model.
No. The licence is CC BY-NC 4.0: free to share and adapt with attribution, commercial use not permitted. That is the one hard constraint in the README and it is stated at the top of the file.
76 seconds for 12 steps and 10 tool calls, 177k tokens split 156k input and 21k output, cross-model verified. It built an offline WebAudio music player end to end, with a streamed answer and live thinking, tool calls with real-time write progress, and finally opening index.html in the browser.
providers/ holds the plugin contract for judgment models and LLMs. context/ carries the chunk store, recursive decomposition, meta-attention, the budgeted context builder and conditional AGENTS.md memory. tools/ injects compact snippets but only the full schemas for tools a judgment pass selected. policy/ runs deterministic rules first, and tracing.py writes every judgment, LLM turn, tool call and policy decision into a per-session JSONL file with cost accounting.
More Jev resources like this
9GitHub stars
Research-memory skill that keeps experiment claims traceable with ten operations, provenance gates, and optional batched Jev audits over memory files.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
poiuyjie/jev_project_contextPython
poiuyjie/jev_project_contextPython11GitHub stars
Local memory layer that stores verbatim agent quotes in per-project SQLite and grades them with Jev into stale, wrong, unsupported, or keep verdicts.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
romiluz13/jevmoryPython
romiluz13/jevmoryPython20GitHub stars
Local harness that runs the open Laya decision model on your own machine and wires it into browser agents through a TypeSafe-compatible /v1/systemone dialect.
Jev Practices & Patterns#community#awesomejev#catalog-import
7GitHub stars
Python client and decision toolkit for the Jev System One endpoint with noul, choice, and score primitives, batched asks, composition helpers, concurrent evaluation, and an MCP server.
Jev Practices & Patterns#community#jevbest#catalog-import
Community
docxology/daf-jevPython
docxology/daf-jevPython7GitHub stars
Independent open-source agent runtime where Jev makes typed tool decisions, an LLM authors text only when needed and reviews low-confidence calls, and code owns execution.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
parkavenue9639/jevloopPython
parkavenue9639/jevloopPython3GitHub stars
Companion code comparing Google ADK and Microsoft Agent Framework on one Pydantic VideoTimeline schema, with the same Jev client checking every generated segment for sane captions and durations.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
jimmyliao/jev-storyboard-labPython
jimmyliao/jev-storyboard-labPythonFrom the Jev guides
Original write-ups on what this entry is for.
Most Jev accuracy problems are state problems. What belongs in a state, what is a pre-made judgment to leave out, and why naming the slice you ask about matters more than the wording.
Jev Practices & Patterns#state#design#primitives
Read article
A Noul returns one number and no separate confidence. How to read the value, pin its boundary with criteria, split compound questions, and set a per-action threshold that matches the cost of being wrong.
Jev Practices & Patterns#noul#primitives#confidence
Read article
A Score is a weighted position, not a ruler. Why the number can land between levels, why levels are judged separately, and how to combine several atomic scores in code instead of one crowded rubric.
Jev Practices & Patterns#score#primitives#scoring
Read article
From the Jev community
Posts from builders shipping with Jev right now.