{"slug": "codeaf", "title": "CodeAF", "summary": "Agent-Field released CodeAF v0.8.3, an Apache-2.0 terminal-based software factory for AI coding agents that run on open models, with 825 GitHub stars and a last commit 3 hours before publication. CodeAF plans a job, runs tasks in parallel on separate branches, and merges only work that passes a held-out acceptance checklist the worker never sees; its \"Pareto Crewing\" assigns a different model to each role and the project claims up to 25× lower cost. The tool works with OpenRouter, DeepSeek, GLM, Kimi, MiniMax, Qwen, Ollama or any OpenAI-compatible endpoint, reads AGENTS.md, CLAUDE.md and existing skills folders, and caps spending with a daily backstop and per-conversation limits.", "body_md": "Open Source Alternative to:\n\n- Stars 825\n- Last commit 3 hours ago\n- Repository age 4 months\n- Version v0.8.3\n- License [Apache-2.0](https://openalternative.co/licenses/apache-2)\n- Repository [agent-field/codeaf](https://github.com/Agent-Field/CodeAF)\n\nThis is a terminal-based software factory for people who want an [AI coding agent](https://openalternative.co/categories/developer-tools/ai-assisted-coding/ai-coding-agents) that works on open models. You describe a job, and it plans the work, runs tasks in parallel on separate branches, and merges only what passes an independent check.\n\nIt ships with three built-in specialists:\n\n- **/senior-dev** implements a coding change and checks the result.\n- **/sec** audits local code for security problems.\n- **/review** reads a GitHub pull request and reports findings, much like an[AI code reviewer](https://openalternative.co/categories/developer-tools/ai-assisted-coding/ai-code-reviewers) . Security and review use read-only tools.\n\nWhat sets it apart:\n\n- **Pareto Crewing** assigns a different model to each role, so expensive models only handle the steps where quality is decided. The project claims up to 25× lower cost.\n- **Held-out acceptance** turns your request into a checklist the worker never sees, so it can't write tests to fit the list.\n- **Rewind** restores full working state, including untracked files like .env, through a bundled tool.\n- **Plain-English rules** cover schedules, hooks and watchers, scoped to a chat, a project or everywhere.\n- **SSH-native** sessions live on your dev box, and you attach from a laptop or phone.\n\nIt works with OpenRouter, DeepSeek, GLM, Kimi, MiniMax, Qwen, Ollama or any OpenAI-compatible endpoint. With Ollama, nothing leaves your machine. It reads your AGENTS.md, CLAUDE.md and existing skills folders, and connects to Slack, Linear, Jira, GitLab, Sentry and others.\n\nSpending is capped by a daily backstop and per-conversation limits. Anything beyond reads and searches asks first. You can use it instead of [Claude Code](https://openalternative.co/alternatives/claude-code) or alongside it, and it's licensed Apache-2.0.\n\nEvery Sunday we deconstruct one proprietary app and pick the best open source alternatives worth switching to.\n\n- Stars\n- Last commit\n- License\n\nAI\n\nAI coding agent for terminal, IDE, and desktop\n\n- Stars Stars growing over the last 30 days228,575\n- Last commit 2 hours ago\n- License MIT\n\nOpen source AI coding agent that works in your terminal, IDE, or desktop app, supporting 75+ LLM providers with no code storage.\n\nCompare with\n\nAI coding agent that works inside your terminal\n\n- Stars 136,140\n\nMinimal terminal coding agent you can fully customize\n\n- Stars 123,322\n\nAutonomous coding agents that complete real engineering tasks\n\n- Stars 94,071\n\nOpen-source AI coding agent for your editor and terminal\n\nCoding agent that runs on local and open-weight models\n\n- Stars 68,944\n\nOpen source AI coding agent for the terminal and IDE\n\n- Stars 30,869\n\nAI coding agent and CLI for complex terminal workflows\n\n- Stars 27,826\n\nAI coding agent for VS Code, JetBrains, CLI, and Cloud\n\nTerminal coding agent with full model and data control\n\n- Stars 1,782\n\nSelf-hosted AI coding agent that opens real PRs\n\nStars growing over the last 30 days\n\nLast commit\n\n2 hours ago\nLicense\n\nApache-2.0\nTerminal-based AI coding agent that inspects files, edits code, runs commands, and integrates into CI pipelines using OpenAI's Codex models.\n\nTerminal-based coding agent with a minimal core, 15+ LLM providers, tree-structured session history, and a TypeScript extension system for building your own workflows.\n\nAI agent platform that runs autonomous coding agents to plan, write, and ship changes across codebases end-to-end, with support for any model and self-hosted deployment.\n\nStars\n\n71,823\n\nAI coding agent that runs inside VS Code and the terminal, supporting multi-file edits, bash execution, and any LLM provider via bring-your-own-key.\n\nCoding agent that lets LLMs write and execute code, manage files, and control your computer using local or low-cost models.\n\nTerminal-based AI coding agent that reads your codebase, edits files and runs commands, with support for Qwen, OpenAI, Anthropic, Gemini and local models.\n\nTerminal-based coding agent powered by Grok 4.5 that plans, builds, tests, and deploys across any codebase with parallel subagents and plugin support.\n\n28,087\n\nOpen source AI coding agent with 500+ models, bring-your-own-key support, and specialized modes for writing, debugging, and planning code across IDEs and CLI.\n\nOpen-source AI coding agent for the terminal that works with 25+ model providers, keeps sessions on your disk, and ships as a single Go binary with no telemetry.\n\nStars growing quickly over the last 30 days350\n\nSource-available, self-hosted AI coding agent that connects to your repo, tools, and chat to investigate issues and open reviewable PRs on your own infrastructure.", "url": "https://wpnews.pro/news/codeaf", "canonical_source": "https://openalternative.co/codeaf?utm_source=openalternative.co&utm_medium=rss", "published_at": "2026-11-22 19:27:51+00:00", "updated_at": "2026-11-22 21:12:30.144232+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "large-language-models", "ai-products"], "entities": ["CodeAF", "Agent-Field", "OpenRouter", "DeepSeek", "GLM", "Kimi", "MiniMax", "Qwen"], "also_reported_by": [], "takeaway": "CodeAF v0.8.3 assigns a different model to each role via \"Pareto Crewing\" and gates merges on a held-out acceptance checklist the worker never sees, which the project claims cuts cost up to 25×.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/codeaf", "markdown": "https://wpnews.pro/news/codeaf.md", "text": "https://wpnews.pro/news/codeaf.txt", "jsonld": "https://wpnews.pro/news/codeaf.jsonld"}}