Search your agent's memory.
GPU Brain gives every agent a durable brain it can search, read, write, and link in seconds — markdown memory through hosted MCP, CLI, and API.
Memory that agents can actually use
A hosted memory layer your agents can search, read, and write to in seconds.
Brain-first lookup
Agents can search memory before answering, so important context is retrieved instead of reconstructed from chat history.
Useful search results
Natural-language recall, exact keyword matches, tags, collections, and page types work together so agents can find the right source fast.
Hosted MCP server
Connect Claude, Cursor, OpenClaw-style agents, and other MCP clients to search, read, write, tag, link, and inspect memory.
Typed links & backlinks
Create relationships between notes, people, companies, and decisions so memory becomes a graph instead of a pile of files.
Markdown memory browser
Humans can browse, edit, and audit the same markdown pages that agents search and update.
CLI and API access
Import folders, create pages, inspect links, run local MCP, or proxy stdio MCP to the hosted endpoint from the terminal.
The agent loop
Read memory, do the work, write back what should survive.
Search first
The agent checks durable memory before it answers, plans, or starts work.
Read sources
It opens the pages, tags, links, and timelines that explain the current situation.
Act with context
Responses, plans, and tool calls use the same memory through MCP, CLI, or REST.
Write back
New decisions, facts, links, and timeline events are saved for the next agent turn.
CLI + hosted MCP
One memory layer for local terminal workflows and remote agents.
# Search your brain
curl -X POST https://gpubrain.app.nz/api/search \
-H "Authorization: Bearer gb_your_key" \
-d '{"query": "notes about fundraising"}'
# Add a note
curl -X POST https://gpubrain.app.nz/api/pages \
-H "Authorization: Bearer gb_your_key" \
-d '{"slug": "ideas/fundraising", "title": "Fundraising notes", "body": "..."}'
# Hosted MCP JSON-RPC
curl -X POST https://gpubrain.app.nz/mcp \
-H "Authorization: Bearer gb_your_key" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
# MCP for Claude, Cursor, or local agents
{
"mcpServers": {
"gpubrain": {
"command": "gpubrain-cli",
"args": ["mcp", "--api-url", "https://gpubrain.app.nz"],
"env": { "GPUBRAIN_API_KEY": "gb_your_key" }
}
}
}
From the blog
How teams use GPU Brain to give their agents memory that lasts.
Instant agent memory
How the read-write memory loop changes everyday agent behavior.
SearchStatic embeddings for memory
Why tiny static models are a good fit for always-on retrieval.
EngineeringGobed as the search engine
How the Go search layer connects markdown, vectors, keywords, and future GPU paths.
Give your agents instant memory
100 free credits on signup. Pro starts at $8 USD/month or $80 USD/year.
Create free account