User repository on GitHub
An open-source merge gateway for repositories where most commits are written by agents.
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Tools & CLIs
Mergeone
An open-source merge gateway for repositories where agents write most commits: a System One model answers five fixed questions per pull request with calibrated probabilities in roughly 300 ms, a YAML policy turns them into pass, hold or block, and a planner lands non-conflicting PRs in waves.
- Stars
- 6
- Language
- Python
- Reading time
- 3 min
- Published by
- Community
- Author
- haalp
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Capabilities
- Open sourceSelf-hosted
Highlights
- System One model answers five fixed questions per PR with calibrated probabilities in one roughly 300 ms call; a policy file turns them into pass/hold/block.
- Ships as one Go binary with a GitHub webhook, HTTP API and MCP server for agents; the README badge claims 11.7 MB and zero runtime.
- Pre-registered gatebench study across 40,000 Kubernetes merges and four systems — two of four runs failed and are reported anyway.
- Badge claims 30M decisions/day at p99 441 ms; the README reports Jev at 100x agent-scale load with p99 417 ms and zero errors.
- Fail-closed policy: model down, timeout or panic holds the PR for a human, and every verdict is logged with its exact input and hash.
Watch out
README reviewed 2026-10-17 from the repository page; the digest copy truncates after the hero section, so this entry draws on the same README read in full. The figures are the author's pre-registered results, several of which failed their declared bars, and the README notes Jev is a closed API in early access whose pricing may change.
Questions about this entry
Short answers drawn from the linked project or page.
An open-source merge gateway for repositories where agents write most of the commits. One typed decision per merge: a decision model scores each pull request on fixed questions, a YAML policy maps the probabilities to pass, hold or block, and a planner lands non-conflicting PRs in waves.
Jev is the decision model behind the gate: it answers five fixed questions per PR with calibrated probabilities in roughly 300 ms, and the README's run 2 reports about $0.074 per 1,000 PRs. Other models, including Claude Haiku, can be swapped in through one YAML row.
The README's 60-second path is a one-line install script, export a TypeSafe API key, copy the generated policy to .mergeone.yaml, then run mergeone pr with base and head refs. Without a key, mergeone serve --mock runs a deterministic stand-in through the whole pipeline.
Run 2 succeeded: typed triage questions scored ROC-AUC 0.87-0.96 against a strong judge. Runs 1 and 3 failed — a diff could not predict 30-day reverts, and the gate could not predict rejection of real agent PRs, scoring ROC-AUC 0.55 against a 0.70 bar.
A README badge claims 30M decisions/day at p99 441 ms, and a load table shows 350 decisions/s on a 2-vCPU box with zero errors using a mock model; with Jev the model itself held 35/s at p99 527 ms against the real API.
No — it fails closed. Model down, timeout or panic means the PR holds for a human, and thresholds such as weakens_tests at or above 0.6 blocking or needs_human at or above 0.5 holding are explicit in the policy file you own. Every verdict is logged with its exact input.
Hasan Alp Temiz (haalp) built it, with Claude as a pair programmer per the README. Metadata: Python, 6 stars, created 2026-10-14 and pushed 2026-10-17; the README's licence badge reads Apache-2.0 and a study site lives at mergeone.dev.
More Jev resources like this
A source-available visual UI testing agent in Go: point it at a URL, describe the test, and a six-stage pipeline drives a real Playwright browser while Jev optionally validates requests, picks DOM targets and checks outcomes, falling back to an LLM below 0.9 confidence.
Jev Tools & CLIs#community#github#go
Community
ownweights/argusGo
ownweights/argusGoLatitude is an open-source observability platform for AI agents: one line of telemetry captures traces, failing traces group into tracked signals, and fixes are dispatched to coding agents and replayed against real traces.
Jev Tools & CLIs#community#catalog-import#logicrw
Community
latitude-dev/latitude-llmTypeScript
latitude-dev/latitude-llmTypeScriptA single-binary gate for Impatient Programming that fails a pull request when an experienced reviewer would ask for the change to be simplified: it asks Jev about each changed file and tells the coding agent what to fix, with no LLM or parser inside.
Jev Tools & CLIs#community#github#go
Community
tools4imps/qualmGo
tools4imps/qualmGoA safety guardrail for coding agents (Claude Code, Codex, Cursor, Antigravity, Grok) that intercepts shell commands, script executions and file writes before they run. Its neural tier can use Liquid AI d1-3B, CLM, TypeSafe Jev cloud or pure rules; a deterministic tier blocks destructive patterns in under 0.1 ms.
Jev Tools & CLIs#community#github#security
Community
Bl4ckk01/agent-guardrailPython
Bl4ckk01/agent-guardrailPythonAn unofficial Go CLI that puts Jev in a coding agent's toolbox: the agent can ask where an answer lives and read only those lines, or hand Jev a yes/no or multiple-choice call and get a calibrated verdict tuned to the cost of a wrong call.
Jev Tools & CLIs#community#github#go
Community
bragamat/jevkitGo
bragamat/jevkitGoAn open-source platform for creating, distributing, and running sovereign EU-compliant LLMs: one Rust binary with OpenAI- and Ollama-compatible APIs, decisions answered as probabilities in milliseconds, and an audit trail designed for the EU AI Act.
Jev Tools & CLIs#community#github#local-llm
Community
eullm/eullmPython
eullm/eullmPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.