There’s a simpler way to ground your agents.
The retrieval pipeline with zero pipes.
Chunking, embedding, indexing, reranking: it all happens inside the engine as data arrives. There’s nothing to wire together and nothing to keep tuned.
# Install and start a local single-node engine, free
curl -fsSL https://releases.antfly.io/antfly/latest/install.sh | sh
antfly standaloneBring whatever data you have.
Antfly sits on your Postgres, your buckets, your file systems, your docs. Nothing to migrate, nothing to replace.
PDFs, screenshots, tables, audio: every context layer becomes food for the retrieval engine.
Vector, keyword, and graph search are indexes inside one database, fused in a single query plan. The inference that chunks, embeds, and reranks runs in there too. That’s where the pipes went.
None of that inference leaves your infrastructure. No third-party API ever sees your data.
Built-in models
Retrieval needs a model at every step: embeddings when documents arrive, a reranker when queries run, an extractor to build the graph. Antfly ships them inside the engine, so there’s nothing to deploy and nothing to call out to.
Built on Antfly
Run it yourself, or let us.
re·triev·al
getting back something that was lost, saved or stored.