aiagentflow
A local-first command-line tool that runs a task through a fixed pipeline of specialized AI agents, architect, coder, reviewer, security, tester, fixer and judge, repeating the loop until quality thresholds pass before shipping.
aiagentflow takes a task, or a project's own specs, PRDs and guidelines, and coordinates a sequence of specialized agents to architect, code, review, test and ship it, accumulating project knowledge in a local memory folder so each run starts smarter than the last. An init wizard auto-detects the project's language, framework, test framework and package manager, lets the user choose from providers including Anthropic, OpenAI, Groq, Gemini, OpenRouter and Ollama, and assigns models per agent role.
An isolate flag runs a task in a clean git worktree so the working directory is left untouched, a review-plan flag pauses after the architect stage so a person can approve, edit or regenerate the plan before coding begins, and separate flags resolve pull request review comments or implement a GitHub issue and open a pull request automatically. A terminal dashboard shows live run status, token counts and per-agent cost, and a batch and parallel mode fan tasks out across multiple simultaneous workflows.
It runs entirely on the user's own machine with no code leaving it, installs from npm, and is written in TypeScript under the MIT license.