jev-fabric
Native Bend executable that owns agent-launched processes and logs, and asks Jev a typed question only when told to.
Overview
jev-fabric is native process orchestration for coding agents, written in Bend with no Node, Python or daemon at runtime. One small executable owns each process an agent starts (dev servers, test runs, game bridges) with its process group, deadline and bounded logs, and remembers what they printed. The agent watches for a ready signal and reads a bounded receipt rather than tens of megabytes of logs.
When one fuzzy judgement remains, the agent can explicitly ask Jev a Choice, Noul or Score question, and the typed answer feeds a branch already written in code, with no hidden inference. It is independent of any particular harness. Installation uses a checksum-verified script for macOS and Linux binaries plus an agent skill, or npm without install scripts; the suggested agent prompt explicitly forbids Jev calls until credentials are given.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Fuzzy judgement about a process's output, asked explicitly by the agent(inferred) | — |
| Yes / no | Did this process reach the expected state?(inferred) | — |
How it works
- Agent starts processes through jev-fabric
- jev-fabric owns process groups, deadlines and bounded logs
- Agent reads bounded receipts and ready signals
- Agent optionally asks Jev a typed question and branches on the answer
Strengths
- Agents avoid drowning in logs
- No runtime dependencies
- Jev calls only when explicitly requested
Caveats
- Early release (v0.1.0)
- Bend is an uncommon language for contributors
Getting started
`curl -fsSL https://raw.githubusercontent.com/fabric-runtime/jev-fabric/main/install.sh | sh` and `npx skills add fabric-runtime/jev-fabric`, then `jev-fabric -- exec /bin/echo ready`.At a glance
- Type
- Service
- Status
- beta
- Language
- Bend
- Runtime
- Native CLI binary with agent skill
- Has code
- yes
Use cases and integrations
How curated lists describe it
Native process orchestration with typed, explicit Jev decisions.
wh000wh000/awesome-jev-live
Frequently asked questions
What does jev-fabric do?
It owns processes an agent starts and gives bounded logs and receipts.
When is Jev called?
Only when the agent explicitly asks a typed question.
What platforms are supported?
macOS universal and Linux x64/arm64 binaries.
Related projects
Jev decision layer for agents: seven MCP tools, an embeddable library and escalate-only Claude Code hooks at harness boundaries.
MCP serverbetaTypeScriptMCP server, CLI and Claude Code skill with Choice, Score and Noul primitives and a confidence gatekeeper, hosted or offline.
MCP serverbetaTypeScriptPortable decision layer for agent harnesses: Jev picks from host-supplied capabilities while the host keeps permissions and execution.
ProjectbetaTypeScriptRust MCP server for Codex that hands classification, relevance, candidate selection and evaluation to Jev with accept/verify policies.
MCP serverbetaRustLocal MCP server putting Jev in the coding loop: route next steps, pick prepared tool calls, review patches and verify claims.
MCP serverbetaTypeScriptDeepSeek Harness plugin that registers a native jev_decide tool and declares its own policy for when the agent should ask Jev.
PluginbetaJavaScript