A local-first knowledge base with semantic search. Built for agents, scripts, and developers who think in pipes.
AI agents are powerful but amnesiac. Tentacle is the missing memory layer.
Find by meaning, not keywords. Your agents stop guessing and start knowing. Query 'auth architecture decisions' and get docs about JWT, OAuth, sessions — even if they never use those exact words.
Every command outputs --json. Built for scripts, pipelines, and agents from day one. No parsing, no scraping, no fragile regex — just clean structured data your agent can consume directly.
No Pinecone. No Docker. No vector database server. Just a single Rust binary and your documents. Local embeddings, SQLite storage, sub-60-second setup. Your laptop is enough.
Why devs switch
| Dimension | Tentacle CLI | Traditional setup |
|---|---|---|
| Setup time | ~30–60 seconds | Hours to days |
| Infrastructure needed | None (single binary + SQLite) | Vector DB, services, credentials |
| Data privacy | Local-first, stays on your machine | Often sent to third-party infra |
| Cost | Free, open-source (optional OpenAI API for auto-tagging) | Usage + infrastructure costs |
Tentacle Pro is a full desktop app — voice capture, RAG chat, and cloud sync. Built for people who think out loud and need to find it later.
Free beta · First 500 users get 3 months of Pro
Tentacle speaks shell. If your agent can run a command or read JSON, it has persistent memory.
#!/bin/bash
# OpenClaw skill: query Tentacle knowledge base
# Called by agent with: $QUERY argument
QUERY="$1"
# Semantic search, return top 3 results as JSON
tentacle search "$QUERY" --limit 3 --json \
| jq '{
results: .results | map({
title: .title,
content: .content,
relevance: .score
})
}'The pattern: tentacle search "…" --json → jq → your LLM
Paste once, run once, and your agent gets persistent memory instantly.
Choose your platform
Free forever. Open source (MIT). No telemetry. No accounts.