semantic search
Semantic search.
Every technical posting gets a semantic embedding, which is how freehire finds postings that mean the same thing in different words — for “similar jobs”, and for anyone who searches by what the work is rather than what it is called. This is how much of the catalogue has one right now — live.
measuring…
of eligible postings embedded
- Embedded
- measuring…
- Eligible
- measuring…
- In the queue
- 0
- Speed
- 0 / min
- Time left
- nothing left
- Queue measured
- 18:56 UTC
“Embedded” is the last full count plus what the queue has drained since, current to the queue’s last sample. The queue also holds closed postings waiting to be cleared, so it can run longer than the eligible count. The full count — measuring… embedded — was last taken at measuring…; it runs every few hours because it reads every open posting.
help speed it up
Lend your machine for a while.
The volunteer client runs the embedding model on your own computer and sends back vectors for postings in the queue. Pick how much of your machine it may use — easy, medium or hard — and watch the queue above move faster.
For AI agents
An agent rarely knows the job title it is looking for. It knows what its user wants to do — “keep payment systems secure”, “build tools for teachers” — and a keyword search misses every posting that says the same thing in other words.
So agents can ask for semantic search: the query is embedded and ranked by meaning as
well as by keywords, over the same embeddings counted above. It is available through the
MCP server, the CLI and the agent API — set semantic to true on a search, and the response says whether
the semantic ranking actually ran. For a plain job title, keywords still answer better.