Last updated 08-22-2026
Category:
Reviews:
Join thousands of AI enthusiasts in the World of AI!
pumaDB
pumaDB gives coding agents and small server-side apps a hosted memory layer without running a database. You get durable JSON tables, a REST API for trusted backend code, and a Streamable HTTP MCP server at https://api.pumadb.ai/mcp that tools like Codex, Claude, and ChatGPT can call directly.
Compared with vector memory services that optimize for semantic search, pumaDB behaves more like a lightweight JSON store built for agent workflows. Rows support revision history with the last 10 archived versions kept for 30 days, atomic batch writes up to 100 operations, and typed remember tools that store code, markdown, and commands as inert reference material rather than executable instructions.
pumaDB fits indie developers, agent builders, and small teams who need persistent task state, user preferences, or research clippings across sessions without provisioning Postgres or Redis. Free accounts include 20 tables, 1000 rows per table, and 25 MB of storage, with Organization Pro raising caps for shared team tables.
Streamable HTTP MCP server at api.pumadb.ai/mcp with OAuth and dynamic client registration
Free tier includes 20 tables, 1000 rows per table, and 25 MB storage
Atomic batch endpoint supports up to 100 operations per request
Keeps the last 10 archived row versions for 30 days with one-click restore
remember tool stores typed memory as code, markdown, command, or config without auto-execution
MCP-first design lets agents read and write memory without custom integration code
Revision history with restore avoids accidental data loss during agent edits
Typed remember tools store code and commands as inert text with safety metadata
Generous free tier covers small agent projects without a credit card
REST and MCP share the same underlying tables for mixed client setups
64 KB row cap on the free tier limits large document storage per record
30 writes per minute per key can bottleneck high-frequency agent loops
Organization Pro pricing is not published on the marketing site
No semantic vector search, so recall depends on explicit table queries and filters
What is pumaDB used for?
pumaDB is a hosted agent memory API for durable JSON tables. Agents and small server-side apps use it to store task state, notes, preferences, research clippings, and app records without provisioning a database.
How do agents connect to pumaDB?
Agent clients should use the hosted MCP URL at https://api.pumadb.ai/mcp with Streamable HTTP transport. Codex users run codex mcp add pumadb --url https://api.pumadb.ai/mcp. The MCP server supports OAuth discovery and dynamic client registration.
What are pumaDB free tier limits?
pumaDB free accounts include 20 tables, 1000 rows per table, 25 MB total storage, and 64 KB per JSON row. API keys are rate limited to 30 writes and 60 reads per minute.
Can pumaDB restore deleted rows?
Yes, pumaDB archives the previous live row on every update or delete and keeps the last 10 versions for 30 days. The restore endpoint recreates a deleted row from an archived version, and version storage does not count against the storage cap.
Should pumaDB API keys go in browser code?
No, pumaDB API keys starting with puma_live_ are full-account bearer secrets for trusted server-side code only. Browser apps should call pumaDB through a backend route that holds PUMADB_API_KEY in server environment variables.
What does Organization Pro add?
pumaDB Organization Pro raises limits on organization-owned tables to 50 tables, 2500 rows per table, 250 MB storage, and 128 KB per row while the subscription is active. Members can create organization tables, and administrators manage billing and roles.