Polygres

What is Polygres

Polygres is a managed PostgreSQL platform that turns your existing Postgres database into internal search for AI agents. It combines rows, relationships, and embeddings into a single hybrid query that returns one ranked answer, making it feel like extended context for AI models.

How to use Polygres

1. Host or connect: Host your database on Polygres, or connect an existing Postgres instance (e.g., Supabase, Neon) using a connection string.
2. Install the CLI: Use pip install polygres-cli to manage projects, import data, and run queries from your terminal.
3. Run a retrieval query: Use the polygres.retrieve() function in your code to query your database. Specify the query, graph hops, filters, and token limit to get a ranked context block.
4. Feed the context to your agent: Pass the retrieved context to your AI agent to ground its responses in your database.

Features of Polygres

• Hybrid retrieval: Combines graph traversal, dense vector search, sparse/full-text search, and more in one query.
• Graph engine (pgGraph): Turns foreign keys into a compiled graph for multi-hop traversal, shortest paths, and neighborhood search.
• AI-native search (pgContext): Runs ten search methods simultaneously and fuses results into one ranked list.
• Dense vector search: Uses HNSW indexes inside Postgres for fast nearest-neighbor search.
• Sparse & full-text search: Catches exact keywords, IDs, and codes that embeddings miss.
• Exact fusion: Merges, deduplicates, and re-ranks results from all channels into one list.
• Unbounded context: Every row in your database is addressable, so agents can access data beyond model context limits.
• Millisecond performance: HNSW and graph retrieval are indexed in Postgres for speed.
• Open source: The underlying engines (pgGraph and pgContext) are open source Postgres extensions.
• CLI: Manage projects, import data, and run queries from the terminal.

Use Cases of Polygres

• AI agent memory: Give agents access to your entire database as working memory, fetching relevant rows on demand.
• Internal search: Turn your Postgres data into a searchable knowledge base for AI-powered applications.
• Customer support: Answer questions like "why did order #8231 fail?" by retrieving related records across tables.
• Recommendation systems: Use graph traversal and vector search to recommend related items.
• Data discovery: Surface relevant rows that agents didn't know to ask for.

FAQ

Q: Can I use my existing Postgres database? A: Yes, you can connect Supabase, Neon, or any Postgres instance with a connection string.

Q: Is Polygres open source? A: The engines under Polygres (pgGraph and pgContext) are open source Postgres extensions. You can run them on your own cluster.

Q: How fast is retrieval? A: Queries return in milliseconds, thanks to HNSW indexing and graph traversal optimized for memory.

Q: Does Polygres require a separate vector store? A: No, embeddings and graph data live inside Postgres, so no sidecar services or sync jobs are needed.