pumaDB vs GBrain

Small durable JSON memory, or a full agent brain?

pumaDB and GBrain both help agents stop losing context, but they sit at different layers. pumaDB is a hosted JSON memory API and MCP server. GBrain is an opinionated knowledge system with retrieval, graph structure, synthesis, ingestion, and background enrichment.

pumaDB

Hosted JSON memory API

GBrain

Retrieval and knowledge brain

Quick read

Choose pumaDB when you want a small hosted memory primitive that an app or agent can start using immediately. Choose GBrain when you want to run and operate a richer personal or company knowledge system.

They are complementary more than direct substitutes: pumaDB is the lightweight memory store; GBrain is the full retrieval-and-reasoning brain.

Best fit

Use the tool that matches the size of the job.

Choose pumaDB for

Hosted MCP memory without running a database
Server-side REST calls from Workers, APIs, CLIs, and scripts
Small structured JSON tables with simple equality filters
Reviewable agent memory with row versions and restore
Typed safe memory for resources, code, Markdown, commands, and config

Choose GBrain for

Markdown knowledge bases synced into a searchable brain
Hybrid retrieval with vectors, keyword search, reranking, and graph signals
Synthesized answers with citations and gap analysis
Typed entity edges such as people, companies, meetings, and relationships
Ingestion recipes, schema packs, skills, evals, and scheduled enrichment

Detailed comparison

Where they differ.

Area pumaDB GBrain

Primary job A small hosted JSON database for agent memory and app state. A full personal or team brain for knowledge retrieval, graph traversal, and synthesis.

Data model Named JSON tables. Current account limits are 20 tables, 1,000 rows per table, 25 MB per account, and 64 KB per row. Markdown pages in a brain repo synced into PGLite or Postgres, with chunks, embeddings, schema packs, and graph edges.

Query behavior CRUD, simple equality filters, sorting, counting, batch writes, and short-lived viewer links for larger results. Hybrid search, vector and keyword ranking, graph-aware retrieval, raw search results, and synthesized answers.

AI layer Stores memory as data. It does not call an LLM to write answers or infer relationships. Positions itself around synthesis, citations, gap analysis, autonomous ingestion, and enrichment workflows.

Operations Hosted on Cloudflare. Connect hosted MCP or call REST from trusted server-side code. Runs locally or self-hosted with PGLite/Postgres/Supabase options and more operational moving parts.

Agent interface Hosted Streamable HTTP MCP, local stdio MCP, REST API, and CLI. CLI, MCP server, agent install protocol, skills, jobs, ingestion recipes, and admin workflows.

Recovery model Every update and delete archives the previous row. The last 10 versions are kept for 30 days and can be restored. Uses a brain repo as a system of record plus database sync, soft deletes, citation repair, and consistency checks.

Positioning

A practical way to think about the split.

pumaDB is the memory primitive

It is useful when an agent needs durable state that stays explicit: preferences, task state, project facts, resource references, code snippets, commands, config examples, and small records your own app already understands.

GBrain is the knowledge system

It is useful when the job is finding, connecting, and summarizing material across a larger corpus of notes, people, companies, meetings, documents, and evolving context.

Sources

What this comparison is based on.

Product pages change over time. This page compares pumaDB's current documented behavior in this codebase with GBrain's public project documentation.

GBrain README pumaDB docs pumaDB install

Start small

Add hosted memory without standing up a database.

Connect pumaDB over hosted MCP, or call the REST API from trusted server-side code.

Install pumaDB Read the API docs