Djævin: calibrated local-first decision layer with a Devin ACP backend - no API key needed
The poor man's Jev. An open source, local-first "System One" decision layer for LLM apps: typed decisions with provably calibrated confidence. No API key. No waitlist.
Community project. Not affiliated with, endorsed by, or sponsored by Cognition AI. "Devin" is a trademark of Cognition AI.
Linux · Personal Windows · Corporate Windows
Part of the awesome-devin ecosystem: the curated hub for the devin- tools.
Fork note: this is the Devin-ecosystem fork of rupeshpoojary9/poordjaevin. It adds a Devin ACP backend: poordjaevin serve scores through the model your Devin CLI already uses, with automatic model rotation and per-call cost telemetry, so Devin users need no extra model download, no Ollama, no VM, and no API key beyond Devin's own credentials. The original keyless local backend remains as the fully offline fallback (POORDJAEVINBACKEND=nli).
Your model's 0.9 is a vibe. poordjaevin's 0.9 is a measurement.
Every LLM-in-JSON-mode hands you a confidence score and hopes you don't check it. poordjaevin checks it. On the shipped eval set it cuts calibration error (ECE) from 0.170 to 0.071 with zero loss of accuracy, and it runs on your laptop with no API key.
Left: raw confidences, overconfident. Right: calibrated, a stated 0.8 really is right about 80% of the time.
[local] pulls torch + transformers for the offline NLI backend. If you only plan to use the Devin ACP backend (poordjaevin serve default), a plain pip install poordjaevin is enough. To install the development version straight from GitHub, append @ git+https://github.com/Icaro0310/poordjaevin.git to the package
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/Icaro0310--poordjaevin.html)
MCP server: turn narrated screen recordings into agent-ready data — local Whisper transcript, scene keyframes, OCR, wall-clock anchoring. Record your screen, talk — your AI agent files the bugs.
PyModel/jev-judge-mcpTyped judgment tools for MCP agents. TypeSafe's Jev model as verify, screen, find, classify, rerank, decide, compare, extract, review, gate, and score: the model judges, policy decides auto, review, o
Shalimov04/mcp-airlockStateless governance proxy for MCP (2026-07-28): policy, dry-run, human confirmation, audit, tracing
ozankasikci/global-agent-memoryLocal-first, project-aware memory MCP server for Claude Code, Codex, and other AI agents, with Obsidian and an owner dashboard.
00200200/usagetrimKeep the signal. Cut the noise. Local CLI + MCP — fold verbose tool output for Claude Code, Codex, Cursor & Desktop.
Riztazz/claude-io-guardA Claude Code plugin that checks every file and shell call an AI agent makes, fixes what it safely can, and tells the agent how to fix the rest. Windows and macOS.
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.