User repository on GitHub
A governed, local-first AI agent for enterprise Chat, Workflows, Associate, and MCP-connected business systems.
- Forks
- 7
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Agents & MCP
Anna-Agent
A local-first desktop agent for macOS that carries work from a conversation through to a reviewable result, split into Home, Cowork and Crew. The part worth a developer's attention is Crew, where Jev suggests an assignee for a single unassigned task and you accept or ignore it.
- Stars
- 147
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- Foxtailsss-Andy
- Last verified
- Oct 2026
About this project
An AI companion for personal tasks, business workflows, and project collaboration. Anna is a personal open-source project exploring how an AI agent can carry work from a conversation through to a reviewable result. The local-first desktop app brings tasks, connected business systems, and collaboration together in Home, Cowork, and Crew. General Agent + Jev Crew · Experimental Source Preview · macOS arm64 · MIT License · CI. 中文 · Product overview · Quick start · Current status · Codex pet · Development diary.
Quoted from the project's README. Read the full README.
- Languages
- TypeScript
Highlights
- Across 24 synthetic heldout cases Jev and DeepSeek both matched all 22 eligible labels, including 10 abstentions, with Jev p50 latency 85% lower.
- Reported total tokens rose 58% on Jev, and the author notes different tokenizers rule out an equivalent-compute reading.
- Assignment suggestions never assign or run anything themselves; you inspect the source and accept explicitly.
- Runs locally with Node, Python and uv, on macOS arm64 only, and the desktop build is unsigned and unnotarized.
Watch out
The evidence is honest about its own size: 24 synthetic cases, two of which never reached a model, and the README states outright that it does not establish production accuracy, Worker completion, or acceleration of the previous complete path. Eight development cases are reported separately, outside that denominator. Cost figures are reference estimates using DeepSeek peak and cache-miss prices rather than invoices, and the lower off-peak estimate is given too. The quick start is not one command: beyond clone, npm ci and uv sync, the harness needs a Bun archive fetched first, and the documented model setup is DeepSeek over an OpenAI-compatible transport, so Jev is not the default provider here. Windows and Linux acceptance is open, authorized write operations against the connected business system are blocked pending a capability the service does not expose, and the Cowork dashboard in the walkthrough runs on synthetic data. 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.
Only one call. The 24 cases measure assignee suggestions for a single unassigned task, and both models got all 22 eligible ones right. The README is explicit that this is not browser end-to-end latency and not a measured speedup of the previous complete Crew/Host/OMP path, and that Worker completion was not measured. The more meaningful comparison is that existing role rules matched only 12 of 22 on the same subset, which is where the decision model is replacing work plain rules already partly handled.
Jev is used for the assignee-suggestion judgment through the production adapter, and a suggested person is never assigned until you accept it in the existing member picker. Everything else runs on the shared Node Harness Host and Oh-my-Pi loop. This piece is labelled a source preview and a candidate, with a release record that separates measured judgments from remaining Worker and platform limits, so treat the integration as scoped and still moving rather than settled.
macOS arm64, Node 22.19 or newer, Python 3.12 and uv. Then npm ci, uv sync with the dev extra, a separate command that downloads a Bun archive for the harness, and npm run desktop:run. The app build is unsigned and unnotarized, so expect Gatekeeper friction. Model access is configured locally and business features additionally need connector configuration. There is no hosted demo described, so you are running the real stack.
No. The connected service used for validation exposed read tools only, and authorized write plus read-back acceptance are recorded as blocked until the service exposes the required capability. What is verified is a browser to Jev to assignment to SQLite readback on synthetic data. Home document generation, prompt creation and Crew worker delivery were exercised earlier against the harness, but the full live provider and MCP gate never closed, so do not plan on live writes.
24 synthetic heldout cases, of which 22 were eligible for model judgment; the other two hard-precheck cases made no model call. Both the DeepSeek and Jev adapters matched the frozen labels on all 22: 12 correct recommendations and 10 correct abstentions, with zero wrong labels or errors.
No. The 85.23% lower p50, 94.71% lower off-peak estimated cost and equal 22/22 label agreement cover model judgments only, and reported total tokens actually rose 57.98%. The author states plainly that this is not a whole-Agent-task speedup claim.
Crew can suggest a person or Worker for one unassigned task, and the picker shows the suggestion's source while keeping Accept assignment separate from generating it. A unique exact-role match uses a rule, Jev handles other eligible semantic choices and can abstain, and suggesting does not assign or run the task.
More Jev resources like this
1.5kGitHub stars
Open-source platform that puts ACP-compatible agents such as Claude Code, Codex, and Pi inside the team's Slack, Telegram, Discord, and GitHub workflows, with reusable Jev decisions for routing, response gating, and reviewer selection.
Jev Agents & MCP#community#jevbest
488GitHub stars
A coding agent built on pi that hands its non-code decisions — what enters the context, whether a command is safe, whether the work is done — to a small judge model at 35 decision points per turn.
Jev Agents & MCP#community#jevbest
Community
qybaihe/muTypeScript
qybaihe/muTypeScript210GitHub stars
An agent plugin for Pi, Claude Code, Codex CLI and Grok Build that asks Jev two classification questions to decide whether a session sits at a safe compaction boundary, then hints or compacts for you.
Jev Agents & MCP#community#jevbest
572GitHub stars
A CLI plus agent hooks that ask Jev one question per rule in your AGENTS.md on every edit or turn and return a probability, so a violating edit gets repaired inside the same turn. Claude Code, Codex, OpenCode, Pi.
Jev Agents & MCP#community#jevbest
Community
coldteadotai/abideTypeScript
coldteadotai/abideTypeScript116GitHub stars
Stanley is a CLI where you describe a code-review request in plain language and Jev routes it to one of ten bounded workflows. Each workflow is deterministic code that gathers bounded evidence and asks Jev fixed-choice questions per hunk; unmatched requests go to the Pi agent.
Jev Agents & MCP#community#jevbest
Community
devagrawal09/jev-codeTypeScript
devagrawal09/jev-codeTypeScript48GitHub stars
Agent harness for work that outlives one chat, with resumable turns, routed state machines, and observational memory across sessions and processes.
Jev Agents & MCP#community#jevbest#catalog-import
Community
dzhng/duet-agentTypeScript
dzhng/duet-agentTypeScriptFrom the Jev community
Posts from builders shipping with Jev right now.