jet
Coding agent where Jev handles context selection, tool routing, plan gating and verification, and the big model sees only key context.
Overview
jet is a coding agent with a native macOS client built on two ideas: Recursive LLM Context Decomposition (RLCD) and judgment-first routing. All context lives as typed chunks in a SQLite tree with no compaction; a fast judgment model decides what each chunk contributes to the next context window (hide, one-line note, summary or full text), so the large model only sees what matters. Jev, or any OpenAI-compatible local model such as laya-mlx, handles context selection, tool routing, plan gating, permission advice and verification.
Small-model answers are disk-cached by content hash so identical decisions are never paid twice; the README cites a run with 17 of 22 cache hits, and a demo that built an offline WebAudio player in 76 s across 12 steps. LLM providers are pluggable (DeepSeek, OpenCode Go, any OpenAI-compatible endpoint), and tools are policy-gated with deterministic rules first and human approval dialogs. It credits a shared design document on a TypeSafe coding agent and is licensed CC BY-NC 4.0, so commercial use is not permitted.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | How should this context chunk appear in the next window? | hide · one-line note · summary · full text |
| Choice | Which tool should handle this step?(inferred) | — |
| Yes / no | Is this step's result verified?(inferred) | yes · no |
How it works
- Store all state as typed chunks in a SQLite tree
- Judgment model decides each chunk's visibility for the next call
- Jev routes tools, gates plans and advises on permissions
- Large model works on the curated context
- Verification gates check results; answers cached by hash
Strengths
- Large model sees far less context, cutting latency
- Cached judgments avoid repeat costs
- Pluggable judgment and LLM providers including local laya-mlx
Caveats
- Non-commercial licence (CC BY-NC 4.0)
- Native client is macOS-only
At a glance
- Type
- Library
- Status
- prototype
- Language
- Python
- Runtime
- agent with native macOS client
- Has code
- yes
- License
- CC-BY-NC-4.0
Use cases and integrations
How curated lists describe it
A TypeSafe-native (Jev) coding agent built on Recursive LLM Context Decomposition (RLCD), with a native macOS client.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide in jet?
Context visibility per chunk, tool routing, plan gating, permission advice and verification.
Can I use a local model instead of Jev?
Yes, any OpenAI-compatible local judgment model such as laya-mlx.
Can I use jet commercially?
No, it is licensed CC BY-NC 4.0.
Related projects
DeepSeek Harness plugin with 12 opt-in Jev or Luna Decisions judgments: skill and file selection, supervision, filtering and approval help.
PluginbetaTypeScriptToolkit where Jev condenses large MCP responses for Claude Code and maps Codex requests to reviewed, fixed repository operations.
ProjectbetaTypeScriptClaude Code plugin with three Jev tools: verbatim context compaction, per-edit model routing and per-prompt skill selection.
ProjectbetaPythonPi extension that strips the inlined skill list and asks Jev which single skill, if any, to load per turn.
Agent skillbetaTypeScriptPairs Jev with the get-fable lifecycle harness to give coding agents fast tool and skill routing and verbatim-preserving compaction.
ProjectprototypeTypeScriptGives an agent one tool that uses Jev to search a nested capability tree and return only the documented commands a task needs.
ProjectprototypeTypeScript