TribalMemory — Your agents don't share a brain. Now they do.

Claude Code

> "Remember: auth uses JWT RS256"

✓ Stored in tribal memory

Codex CLI

> "How does auth work?"

Based on tribal memory:

Auth service uses JWT with RS256 signing...

tribal_store → one agent learns · tribal_recall → every agent knows

FEATURES

--semantic-searchFind memories by meaning, not keywords

--cross-agentMemories from one agent → all agents

--graph-searchEntity extraction + relationship traversal

--hybrid-retrievalVector + BM25 keyword search combined

--local-firstFastEmbed ONNX: zero cloud, zero API keys

--session-indexingIndex conversation transcripts for search

--deduplicationWon't store the same thing twice

--temporal-reasoningDate extraction and time-based filtering

--import-exportPortable JSON bundles across instances

--mcp-nativeWorks with Claude Code, Codex, and more

--auto-captureSaves memories as agents respond

--/rememberInstant save: /remember <thing>

GET STARTED

$ pip install tribalmemory

$ tribalmemory init

$ tribalmemory serve

That's it. Server running. Memory shared. No config needed.

# Want OpenAI embeddings instead?
$ tribalmemory init --openai

# Already running Ollama?
$ tribalmemory init --ollama

CLAUDE CODE

{ "mcpServers": { "tribal-memory": { "command": "tribalmemory-mcp" } } }▌

CODEX CLI

[mcp_servers. tribal-memory] command = "tribalmemory-mcp" # ~/.codex/config.toml▌

OPENCLAW

$ openclaw plugins install ./extensions/ memory-tribal▌

LOCAL MODE = ZERO DATA LEAVES YOUR MACHINE

▸ Embeddings computed locally (ONNX runtime)

▸ Memories stored locally (LanceDB)

▸ No API keys. No cloud. No telemetry.