Dedicated Postgres per project
Every project gets its own database and compute. No shared logical databases.
Trusted by F1000 enterprises
The AI-Native Supabase Alternative RAG and Agents built-in
Powabase is the Postgres backend for AI apps. Every project gets its own database, auth, storage, and dedicated compute. Documents index on upload, agents call tools over HTTP or MCP, and all of it sits behind one REST API.
Trusted by teams at
CODING AGENTS & VIBE TOOLS
Point your coding agent at Powabase.
Whatever you build with, Powabase is the backend behind it. Connect through the agent skill, the REST API or the auth guide, and build on real Postgres, storage, auth and agents.
Claude Code
AnthropicAnthropic's agentic coding tool for the terminal, IDE, and CI.
SkillAPIAuth
Codex
OpenAIOpenAI's CLI agent with sandboxed file and shell access.
SkillAPIAuth
Antigravity
GoogleGoogle's Gemini agent IDE that runs agents in parallel.
SkillAPIAuth
OpenCode
AnomalyOpen-source terminal agent that runs on 75+ model providers.
SkillAPIAuth
Replit
ReplitCloud IDE and Agent that builds and hosts full-stack apps.
APISkillAuth
Lovable
LovableTurns a prompt into a working web app.
APIAuth
v0
VercelVercel's tool for generating and shipping Next.js UI.
APIAuth
Base44
Base44Builds complete internal tools straight from a prompt.
APIAuth
Bolt.new
StackBlitzIn-browser builder that writes, runs, and deploys apps.
APIAuth
BUILD WITH AI
Your coding agent builds the backend.
Install the Powabase skill or connect the Powabase MCP server, and your coding agent speaks the platform natively. Describe what you want and it wires up retrieval, agents, and workflows behind a single REST API.
support-bot · claude
$ npx skills add powabase-ai/agent-skills
✓ powabase skill installed · your agent now speaks Powabase
› Build a support agent that answers from our product docs.
connecting to project support-bot · app.powabase.ai
✓Sources200 PDFs → text + chunks
✓Knowledge baseembedded & indexed in pgvector
✓Agentclaude-sonnet · knowledge_search
✓Live APIPOST /agents/run/stream · SSE
✦Live on your site→ yoursite.com/support· powered by your Powabase backend
One REST API instead of a vector database, an agent framework, and the glue code between them.
Retrieval built in
Documents extract, embed, and index on upload. Vector, full-text, and hybrid search over your own data.
Agents that take action
Tools over HTTP or MCP, with persistent sessions, streaming, and human approval for sensitive calls.
BACKEND
The BaaS part.
Postgres with RLS, a first-class auth service, object storage, and realtime. Access it through PostgREST, the REST/GraphQL-style API, or a direct database connection.
RETRIEVAL
RAG, out of the box.
Upload PDFs, images, office files, or URLs and Powabase extracts, chunks, embeds, and indexes them. Our built-in OCR hits 91% accuracy on OlmOCR-Bench, and the full RAG pipeline scores 98.7% on FinanceBench. BM25, pgvector, hybrid search, and SOTA rerankers come included, and multimodal content gets indexed too.
AGENTS
Agents, with tools.
Define ReAct orchestrations with multiple LLMs, knowledge bases, and tools. Runs stream over SSE, and we log every retrieval event, tool call, token delta, and citation. Sessions live in your project's Postgres. When a conversation outgrows the model's context window, the agent summarizes older turns and keeps the recent ones word for word, so a long session never loses the thread. Start with built-in tools like web search and code execution, then add your own over HTTP or MCP.
WORKFLOWS
Workflows, visual and callable.
Build multi-step agent workflows by dragging and connecting blocks: triggers, conditions, agents, HTTP calls, code. Ask the natural-language copilot to design the flow for you. Deploy a workflow and turn it into an HTTP endpoint.
ARCHITECTURE
Your own stack. Tuned for AI.
Every Powabase project gets a fully isolated stack with its own Postgres, Realtime, and Storage. Nothing sits in a shared logical database, so there's no noisy-neighbor risk and your SOC 2 / ISO 27001 compliance assumptions hold by default. We built the compute underneath for AI workloads. Retrieval, rerank, and the agent runtime run side by side, which keeps RAG hot and agent loops short.
ISOLATION
Isolated Postgres + Realtime + Storage
per project
No shared logical databases
no noisy-neighbor risk
Compliance-by-default
SOC 2 · ISO 27001
AI-TUNED COMPUTE
Co-located retrieval, rerank & LLM gateway
no extra network hops
Hot vector index next to Postgres
sub-100ms lookups¹
Up to 70% token savings
on optimized agents
¹ Warm-cache pgvector lookups on typical RAG payloads.
OPEN SOURCE
Full open source. Self-host in one command.
The whole stack lives in one repo under Apache-2.0: Postgres, auth, storage, REST, and the Powabase AI service. Clone it, run docker compose up, and you have a single-project Powabase on your own infrastructure. No control plane, no account required.
- Apache-2.0
- Single docker compose up
- 12 services, no control plane
powabase · self-host
git clone https://github.com/powabase-ai/powabase
cd powabase
docker compose up
12 services up · Studio on http://localhost:3000
DEPLOYMENT
Run it your way.
Ship on our managed cloud, or self-host the open-source stack yourself. Either way, bring your own LLM keys so costs and compliance stay with you.
MANAGED
Powabase Cloud
Fully managed and provisioned in seconds. We handle updates, backups, and scaling for you.
OPEN SOURCE
Docker Compose
Self-host the full single-project stack with one docker compose up. Apache-2.0, free to run anywhere.
ENTERPRISE
Kubernetes
Scalable, high-availability deployment on your own cluster via the official Helm chart, with support and SLAs.
Helm chart · Enterprise
Powabase is open source under Apache-2.0. Self-host the full stack for free; Enterprise adds a supported edition with SLAs and private hosting. See Enterprise →
COMMUNITY
Join the community
Support inquiries, feature requests, and show off your apps.
Shipping something serious? Apply for a free MVP build.
FROM THE BLOG
Notes from building the backend.
How retrieval, agents, and multi-tenant isolation actually work on Postgres.
Agents & Workflows Agentic AI Trading Strategy: A Step-by-Step Guide
Master an agentic AI trading strategy with our clear, step-by-step guide. Learn how to build, test, and deploy autonomous trading systems that act on your
RAG & Retrieval How to Make Multitenant Vector Indices Scalable
Multitenant vector indices can bottleneck fast-growing apps. Learn proven strategies to scale them efficiently without sacrificing speed or isolation.
Agents & Workflows How to Build a Hedge Fund Trading Model with Agentic AI
Learn how to build a hedge fund trading model with agentic AI: design multi-agent roles, wire data APIs, manage risk, backtest, and deploy to paper trading.
Agents & Workflows Prompt Caching for AI Agents: Where the Savings Come From
Learn how prompt caching for AI agents cuts cost and latency in long loops — KV cache basics, provider differences, cache hit rates, and best practices.
Backend & Comparisons Row-Level Security for AI Agents: Run as User or Not?
Row-level security for AI agents raises a key question: should they run as the user or not? Explore the tradeoffs to make the right call for your system.
Backend & Comparisons Claude Workspace vs Custom Build: When to Build
Claude workspace vs custom build: learn where Claude Projects are enough, where they hit a wall, and when to build custom with Powabase.
FAQ
Questions.
Powabase. It's the Postgres backend for AI apps: per-project Postgres with pgvector, a built-in RAG pipeline, and an agent runtime behind a single REST API.
Powabase is the Postgres backend for AI apps. Every project gets its own Postgres, auth, storage, realtime, and dedicated compute, plus a RAG pipeline, an agent runtime, and workflows behind one REST API, so you never have to bolt a vector database and an agent framework onto your backend. Coding agents like Claude Code, Codex, and Cursor drive it through the Powabase Agent Skill or the Powabase MCP server.
Powabase builds on the Supabase stack, so you start with the same Postgres, auth, storage, realtime, and PostgREST. On top of that, we ship a RAG pipeline, an agent runtime, orchestrations, and workflows in the same project, with dedicated compute per project. If you only need a database and auth, Supabase covers it. If you'd otherwise run a vector database and an agent framework next to it, Powabase replaces that stack with one API.
Yes. Supabase ships pgvector, so you can store embeddings and run similarity search yourself. Extraction, chunking, embedding, indexing, reranking, and the agent layer are then yours to build. Powabase ships that pipeline on the same Postgres. Documents extract, embed, and index on upload, with vector, full-text, hybrid, and tree search built in.
No. Embeddings live in pgvector inside each project's own Postgres, next to your app data, so there's no separate vector database to run, pay for, or keep in sync.
Yes. Every Powabase project is a dedicated stack with its own Postgres, storage, keys, and compute. There are no shared logical databases, no cross-tenant traffic, and no noisy neighbours.
A production Postgres database with instant REST APIs (PostgREST), GoTrue authentication, object storage, realtime, and Row-Level Security. The AI layer sits in the same project: knowledge bases (RAG), agents, multi-agent orchestrations, and workflows, all on one consistent API.
Upload PDFs, Office files, images, or URLs and Powabase extracts (with OCR and vision models for scanned or image-heavy files), indexes, and retrieves them automatically. Embeddings live in your own Postgres (pgvector), with vector, keyword, and hybrid search plus reranking and multimodal retrieval. No separate vector database to run.
PDFs (including scanned, via OCR), Word, PowerPoint, and Excel, plain text, Markdown, and CSV, images (PNG, JPG, WebP, GIF, TIFF), and web pages crawled from URLs.
A model of your choice plus tools, optional knowledge bases, and session memory, running a reason-act-observe loop. Agents get eight built-in tools (database read/write, HTTP, code execution, storage, web search and scrape), plus your own HTTP tools and MCP servers, with human-in-the-loop approval for sensitive actions. They compose into orchestrations and workflows.
Within a session, yes. Every run is stored in your project's Postgres and reloaded when you continue the session, and sessions last until you delete them. When a conversation outgrows the model's context window, the agent summarizes older turns and keeps recent ones word for word, while the full transcript stays in Postgres. Memory doesn't carry across separate sessions on its own; for that, have the agent write to a table or knowledge base.
Bring your own keys for OpenAI, Anthropic, Google, and OpenRouter (OpenRouter also unlocks Mistral and open-source models). API keys are stored per-project and encrypted at rest. Don't want to manage keys? Skip them and pay for inference from your Powabase credits instead.
Powabase uses usage-based pricing, not per-seat. You pick a monthly plan whose fee converts into a pool of credits you spend on actual usage (compute hours, per-call actions, active users, and storage), and unused credits roll over month to month. Free is $0/month (full platform, pay-as-you-go at standard rates); Self Serve ($25/mo) and Scale ($300/mo) each include that amount in monthly credits plus progressively cheaper unit rates; Enterprise is custom. Moving up a plan buys prepaid credits and cheaper usage, never seats.
Two models: Powabase Cloud (fully managed at app.powabase.ai) and self-hosting the open-source stack yourself. The self-hostable edition is a single-project stack you run with one docker compose up (Apache-2.0, at github.com/powabase-ai/powabase). Enterprise adds a commercially-licensed, supported edition with SLAs, a Kubernetes Helm chart, and private managed hosting.
Yes. Powabase is open source (Apache-2.0). Clone github.com/powabase-ai/powabase and run docker compose up to get the full single-project stack on your own infrastructure.
Yes. It's Apache-2.0 and lives at github.com/powabase-ai/powabase. The self-hostable edition is a single-project stack (Postgres, auth, storage, REST, plus the Powabase AI service) you run with one docker compose up.
Each project is an isolated stack with no shared database, per-project keys, and Row-Level Security keeping data separated. Enterprise plans support SOC 2 and ISO 27001, with DPA and SLA terms available.
Every feature is an API, runs stream over SSE, and tools speak HTTP or MCP. Our docs are machine-readable for coding agents like Claude Code or Codex, and the Powabase MCP server lets any MCP client query your database and build agents, knowledge bases, and workflows directly.
Claude Code, Codex, Antigravity, OpenCode, Replit, Lovable, v0, Base44, and Bolt.new. Coding agents install the Powabase Agent Skill (npx skills add powabase-ai/agent-skills) or connect the Powabase MCP server (https://mcp.powabase.ai/mcp) and build the backend from a prompt; browser-based platforms connect over the REST API using your project's Service Role key from the Connect dialog (kept server-side). See powabase.ai/integrations.