
PostgresML
Not yet source-checkedPostgresML is a Postgres extension that combines data storage and machine learning inference within the database.
Machine-drafted from the vendor's site · not yet source-checkedProfile updated Link checked Pricing captured How we rate
In short
PostgresML is a Postgres extension that unifies data storage and ML inference. It enables RAG and vector search using SQL or SDKs within a single GPU-backed database.
Organizations with established workloads.
PostgresML keeps vectors, models, and application data in one GPU-backed database to ship RAG and ML features with fewer moving parts. Users can interact with the system using SQL or SDKs in JavaScript and Python. The vendor says that the serverless cloud provides a free database with access to GPUs and state-of-the-art LLMs for getting started. Users can self-host the extension if they do not want to use the cloud service. The tool transforms a PostgreSQL database into a vector database for Retrieval-Augmented Generation applications.
According to PostgresML, PostgresML keeps vectors, models, and application data in one GPU-backed database to ship RAG and ML features with fewer moving parts. Its pages list this strength: Runs machine learning and AI operations directly within PostgreSQL. They also state this limitation: PostgresML does not directly support integration with remote LLM providers like OpenAI.
— The AI Tool Bible editorial team
Pros
- ✅ Runs machine learning and AI operations directly within PostgreSQL.
- ✅ Leverages GPU power for faster computations and model inference.
- ✅ Includes built-in functions for chunking, embedding, ranking, and transforming text.
- ✅ Provides efficient similarity search using pgvector integration.
- ✅ Offers more than 47 classification and regression algorithms.
Cons
- ⚠️ PostgresML does not directly support integration with remote LLM providers like OpenAI.
Pricing (captured )
| Tier | Monthly | Key features |
|---|---|---|
| Serverless | From $7.50 per query hour |
|
| Dedicated | From $0.60 per instance hour |
|
| Enterprise | Custom pricing |
|
Prices change; we show the grid as captured on Oct 12, 2026. Check current pricing on the vendor's site →
Facts
- Platforms
- api
- API available
- Yes
- Open source
- Yes (license: MIT)
- GitHub stars
- 6,832 (checked Nov 13, 2026)
- Last GitHub push
- Jul 1, 2025
- First commit
- Apr 2022
- Company
- PostgresML
- Model used
- Multi-model (Llama, Mistral, open-source embeddings)
- Hacker News
- 54 stories · top 444 points · last Oct 9, 2024 · checked Nov 19, 2026
- Earliest Wayback capture
- 2022-05-14 (earliest archived capture of the site, not a launch date)
Where this tool sits
Use cases
Frequently asked
- How much does PostgresML cost?
- PostgresML offers a freemium model. Serverless plans start at $7.50 per query hour, while dedicated instances start at $0.60 per instance hour. The serverless cloud also provides a free database with GPU access for getting started.
- Which machine learning models are supported?
- The tool supports a multi-model approach, including Llama, Mistral, and open-source embeddings. This allows users to leverage state-of-the-art LLMs and embedding models directly within their database environment for various ML tasks.
- Can I self-host PostgresML instead of using the cloud?
- Yes, users can self-host the extension if they prefer not to use the cloud service. This flexibility allows organizations to maintain control over their infrastructure while still utilizing the tool’s capabilities for vector search and ML inference.
- How do I interact with PostgresML?
- Users can interact with the system using standard SQL queries or through SDKs available in JavaScript and Python. This dual approach makes it accessible for both database administrators and application developers integrating ML features.
- What is PostgresML best suited for?
- It is best for organizations with established workloads that want to reduce moving parts. It transforms PostgreSQL into a vector database, ideal for Retrieval-Augmented Generation, vector search, embedding generation, and model fine-tuning.