Intake
Turn messy requirements into verified specs for AI agents
3 followers
Turn messy requirements into verified specs for AI agents
3 followers
Intake is an open-source CLI that ingests requirements from any format (Jira, PDFs, Confluence, images, Markdown, YAML...), analyzes them with AI, and generates executable specs that any coding agent can use. Multi-source. Model-agnostic. Verifiable.
I'm Diego, and I built intake to solve a problem every developer using AI coding agents hits eventually: the gap between scattered requirements and the structured spec your agent actually needs.
The problem is simple. Your requirements live everywhere: Gitlab, Jira tickets, PDF docs, Confluence pages, random Markdown files, etc. Before any AI agent (Claude Code, Cursor, Copilot, Architect, Kiro) can start coding, someone has to read all of that, synthesize it, and translate it into something the agent can work with. That someone is usually you, manually, every single time. intake automates that entire step.
Here's what it does:
🔌 Ingests anything --> Markdown, PDF, DOCX, JSON, Jira, Confluence, GitLab Issues. Point it at your sources and go.
🧠 AI-powered analysis --> Extracts requirements, detects conflicts between sources, deduplicates, identifies risks, and flags open questions. Uses any LLM you want (Claude, GPT, Gemini, DeepSeek, local models via LiteLLM). Boost with LiteLLM.
📋 Generates 6 spec files --> requirements.md, design.md, tasks.md, acceptance.yaml, context.md, sources.md. Everything an agent needs, with full traceability back to the original sources.
✅ Executable verification --> intake verify runs acceptance checks against your actual codebase. Works in CI/CD. Exit codes, JUnit XML, the works.
🔄 Agent-agnostic export --> Output formatted for Claude Code, Cursor, Kiro, Copilot, or any custom agent. intake doesn't replace your agent, it feeds it.
🧩 Plugin system --> Parsers, exporters, and connectors are pip-installable plugins. Extend intake without forking.
📡 MCP Server --> Agents can consume specs in real time during development sessions via the Model Context Protocol.
What makes intake different from tools like Kiro or GitHub Spec Kit? Those tools accept a single prompt as input. intake accepts N sources in N formats, merges them into one coherent spec, and detects what contradicts what. That's the gap it fills.
intake is fully open source (MIT), installs with pip install intake-ai-cli, and works on Linux, macOS, and Windows. Python 3.12+.
I'd love your feedback,especially on which integrations or exporters to prioritize next. Star the repo if this is useful to you!