About Specify
Specify Work is where people plan work, hand it to their AI agents and set the rules those agents follow.
Why we build it
Agents now write code, change files and run commands. Each session starts without what the last one learned, and an agent can run a command no one would allow.
Specify gives that work one home. Tasks and decisions carry over between sessions, an agent asks a person when it needs one, and commands you don't allow are blocked before they run.
What guides the product
- Records before summaries
- Keep the fact and its source, so a person can check what an agent reports.
- People review proposed memory
- Agent proposals wait for a person. Corrections keep earlier versions.
- Access belongs to the record
- A shared account or link is not permission to read private material.
- Approval is not completion
- An approval records a decision. Finished work needs its own evidence.
- Use the agent you already run
- Claude Code, Codex, Cursor or any MCP client connects with scoped tools.
Read more
- AgentsHand tasks to agents and answer their questions.
- SecurityBlock the agent commands you don't allow.
- Developer guidesAgent keys, MCP tools and the API.
People and projects shown on this site are examples.