AI agents are brilliant at reasoning. But they burn tokens re-solving the same problems — extracting PDFs, calling APIs, formatting code — problems that someone else already solved.
We built Dejavu to fix this. It's a marketplace where agents find and use pre-built skills instead of reasoning from scratch. Here's what we learned.
Every AI agent user has the same experience: your agent spends 30 seconds "figuring out" how to read a CSV file. It imports three libraries, tries two approaches, handles edge cases. It works. Then it does the exact same thing tomorrow.
Someone already solved this problem. Their solution is tested, debugged, and sitting in a public repo. The agent just can't find it.
The insight: don't make agents smarter. Make them better at reusing what already works.
Dejavu gives AI agents the ability to discover and use pre-built skills — no human browsing required. Install once, and your agent gains the ability to search thousands of verified skills and execute them instantly.
You don't change your workflow. Your agent doesn't change its reasoning. It just has a new tool: when it hits a problem, it checks if a proven solution already exists. Most of the time, it does.
Most skill directories are searchable catalogs for people to browse. That's useful, but it's the wrong interface. Agents don't browse — they query.
We built Dejavu so the agent does the discovery. It searches, finds the best match, loads it, and uses it. The human just sees faster, cheaper results. No new workflow to learn.
Your agent shouldn't wait for a network call every time it wants to search for a skill. Skills should be available offline, with sub-millisecond lookups. The catalog lives on your machine — sync once, search forever.
There are directories with hundreds of thousands of entries. Most are raw scrapes with no quality filtering — broken repos, abandoned projects, README files masquerading as skills.
We require that every skill in Dejavu is both documented and executable. That filters out most candidates, but it means every result is actually useful.
We chose $6.67/month flat. No per-use metering. No credit systems. No free tier that turns into a paid one later.
Why? Because agents should be able to search and use skills freely without worrying about a meter running. Flat pricing aligns our incentives: we succeed when you keep using it, not when you use it more expensively.
Creators earn 70% of revenue from their skills. Every subscriber is a potential creator — when your agent solves a problem, that solution can become a skill that earns money.
We're growing the catalog, improving discovery, and making the agent experience seamless. The goal isn't more features — it's making the existing ones disappear into the background. Your agent should use Dejavu without you thinking about it.
pip install dejavu-mcp dejavu-mcp serve --api-key your_key
Your agent now has skill discovery. First search is instant. First execution downloads the skill. Every subsequent execution costs nothing.