Quipu

io.github.scbrownPublished by GitHub account scbrownv0.15.0Updated Oct 24, 2026

Local knowledge graph memory for agents. SHACL rules refuse bad facts; SPARQL and time travel.

VerifiedSTDIODesktop onlyDatabasesKnowledge & MemoryDeveloper Tools

Overview

AI-generated overview

Quipu is a local, governed knowledge graph memory for AI agents, letting an assistant store facts with provenance and query current or historical state.

What you can do
  • Store facts written in Turtle into a local graph database, with writes checked against your vocabulary and shapes
  • Query the graph with SPARQL, including earlier versions of the data for time-travel questions
  • Search stored knowledge by keyword, or semantically once an embedding model is configured
  • Import verified knowledge bundles that stage new facts without changing the existing root
  • Share project facts, decisions and relationships across agents and tools instead of repeating context
When to use it
Useful when you want an assistant to keep durable, auditable project knowledge — facts, decisions and their history — in one local store that several agents and tools can read and reuse, rather than pasting the same context into each session.
Requirements
Runs as a local process over stdio, started with a database path, for example via the Docker image or the command-line tool. The CLI targets Linux and macOS on x86_64 and arm64 and needs Python 3.9+ and curl for the installer. No environment variables, headers or authentication are declared. Semantic search needs an embedding model installed; keyword search works on a fresh store.
Before you install
The server writes to and reads from a local database file, so point it at the store you intend and back it up. Imports and writes change stored knowledge, though imports are described as staging without altering the existing root. Semantic search requires downloading a model. The README's install step pipes a remote script into a shell, which should be reviewed before running.

Installation

In SourceWeft

  1. Open Quipu in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

[Quipu logo: knotted strings forming a knowledge graph]

Quipu

Governed knowledge your agents can read, reuse and share

[License: Apache-2.0] [CI] [crates.io: quipu-ai] [Docs: the Quipu book] [DOI: 10.5281/zenodo.21878428] [arXiv: 2608.16813] [Part of the Tambo stack]

Read the book →

Quipu is a governed, bitemporal knowledge store for AI agents and the people who run them. Use it from the command line, REST, MCP or a Rust application.

Why you would want it

  • Governed: check incoming knowledge against your vocabulary and shapes.
  • Bitemporal: keep a queryable history of facts and decisions.
  • Reusable agent knowledge: store memory agents can read, reuse and share across tools.
  • Time-aware entities: attach event-triggering time attributes for Chaski to react to.

Use Quipu as the central graph and base for many tools: see the stack. The full case is in Why Quipu.

[Animated: one Quipu store connects the CLI, MCP clients, HTTP server, Rust library, offline use and WebAssembly]

Install

sh
curl -fsSL https://raw.githubusercontent.com/intigraph/quipu/main/install.sh | sh

Linux and macOS, x86_64 and arm64; requires Python 3.9+ and curl. The installer verifies the published archive checksum and executable version. Then run quipu --version. Install, upgrade or remove Quipu. For the whole stack, use Tambo's guided onboarding.

First success: insert a fact, watch the graph change

Write one fact in plain Turtle, knot it into a local store, and ask for it back:

sh
printf '@prefix : <urn:demo:> .\n:Ada :studies :Engine .\n' > ada.ttlquipu knot ada.ttl --db demo.dbquipu read 'SELECT ?who ?what WHERE { ?who <urn:demo:studies> ?what }' --db demo.db

Add a second fact and the graph grows. The first version is still there to query:

sh
printf '@prefix : <urn:demo:> .\n:Babbage :designed :Engine .\n' > babbage.ttlquipu knot babbage.ttl --db demo.dbquipu read 'SELECT ?s ?p ?o WHERE { ?s ?p ?o }' --db demo.dbquipu read 'SELECT ?s ?p ?o WHERE { ?s ?p ?o }' --tx 1 --db demo.db

[Animated terminal: two plain facts are inserted and queried, then the earlier version is read back]

After the first factAfter the second fact
[Graph before: Ada studies Engine][Graph after: Ada studies Engine, and Babbage designed Engine]

Transcript, tool version and reproducible capture.

On your own code

Keep project facts, decisions and relationships together instead of pasting the same context into each agent.

You want toStart here
Load facts and apply your vocabularyQuick Start
Ask about current or historical knowledgeSPARQL and time
Share a governed knowledge bundleVerified sharing
Prepare domain knowledge and shapesCamayoc

[Animated: export a share, verify it in quarantine, then explicitly promote admitted knowledge; conflicts are not silently merged]

The latest release includes a repository pendant, checksum and provenance. Verify them together, then use quipu import SHARE_DIR --db repository.db; import stages knowledge and leaves ROOT unchanged. Follow the verified import workflow.

Wire it into your agent

For Claude Code, connect the local MCP server to your database:

sh
claude mcp add quipu -- quipu mcp --db /path/to/knowledge.db

The MCP Registry identifier is mcp-name: io.github.intigraph/quipu. Other clients, HTTP MCP and credential configuration: Agent Setup. Use Tambo to configure and verify the tools together.

Before you start

  • The CLI runs locally; quipu-server supplies the network API and explorer.
  • Fresh stores support keyword search without downloading a model. quipu search "your terms" uses keyword search until an embedding provider is configured; explicit --mode semantic still requires a model and vectors. Install the pinned default model with quipu model install in a full build. Check readiness with quipu doctor search.
  • Configuration covers storage and server settings.

What's next

🧺 The stack

Tambo installs these together and proves each one works; every tool also stands alone. Use each tool's README for its own installation and first success.

The Quipu stack is built around a governed, bitemporal knowledge store: shared facts, decisions and work history your agents can query. Quipu can check writes against your rules, records where each fact came from, and answers what was true when.

toolwhat it gives your agents
tambo🧰 Agent Platform Toolkit - installs and configures the quipu stack
quipustores shared facts with provenance and history; can check writes against your rules
camayocprepares knowledge and vocabulary for Quipu, with checks before it is admitted
quechuashared vocabulary and shapes for interoperable knowledge
chaskireacts to graph changes and delivers governed notifications
bobbinindexes your codebase and injects context from that index and Quipu knowledge
yupanashows code impact and can read policies from Quipu to govern agent actions
desire-pathturns agent mistakes into feature requests; works on its own without Quipu
seedsa work item tracker backed by the Quipu knowledge base
shantytownruns crews of AI agents on work items, triggered by hand, on a schedule or by events; can share work and memory through Quipu
rikuylive brief: status, graphs, watches and signed decisions over Quipu
shuttleruns signed agent workflows and keeps their records in Quipu

Contributing

See CONTRIBUTING.md. The repository's checks are:

sh
just checkjust docs check

📜 License

Apache-2.0.

Source: README.md at commit 2289024

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

10
  1. v0.15.0LatestOct 20, 2026
  2. v0.14.1Oct 15, 2026
  3. v0.14.0Oct 15, 2026
  4. v0.13.0Oct 14, 2026
  5. v0.12.0Oct 10, 2026
  6. v0.11.1Oct 8, 2026
  7. v0.11.0Oct 3, 2026
  8. v0.10.0Oct 2, 2026
  9. v0.9.1Sep 30, 2026
  10. v0.9.0Sep 26, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts