LibreAI / KI-Agenten / honcho
// agents · plastic-labs/honcho

honcho

honcho ist ein Open-Source-Projekt aus der Kategorie KI-Agenten von plastic-labs (plastic-labs/honcho auf GitHub). Es hat 7.536 Sterne auf GitHub und 928 Forks, wird unter der Lizenz AGPL-3.0 veröffentlicht und wurde zuletzt am 8. Oktober 2026 auf GitHub aktualisiert.

Honcho — Plastic Labs' memory library for stateful agents that learn about their users over time.

★ 7,5 Tsd. Sterne⑂ 928 ForksAGPL-3.0aktualisiert Okt 2026
▦ Visual — das README dieses Projekts enthält keinen einbettbaren Screenshot,
darum hier ein Platzhalter. Die Anleitung unten ist trotzdem vollständig.
// projektdaten

Projektdaten

GitHub-Sterne
7.536
Forks
928
Lizenz
AGPL-3.0
Zuletzt auf GitHub aktualisiert
8. Oktober 2026
Kategorie
KI-Agenten
Eintragstyp
Vollständige Installationsanleitung
Offizielle Website
Official site
// so funktioniert es

So funktioniert es

  1. Clarify your goal

    Decide managed versus self-hosted, and model your world as peers and sessions — typically your users and your agent as peers exchanging messages.

  2. Get the project

    Clone plastic-labs/honcho from GitHub (or use the install path in the run steps below) and skim its README for the parts relevant to your goal.

  3. Install and configure

    Install the SDK for your language; for self-hosting, clone the repo and start it with Docker Compose, then point the SDK at your local base URL.

  4. Run your first task

    Store a few messages on a session, let background reasoning run, then query Honcho for what it knows about that peer and inject the context into a prompt.

// funktionen

Warum es so beliebt ist

The Honcho loop

Store messages, let asynchronous background reasoning build understanding, query it, and inject prompt-ready context into any model.

Peers and sessions model

Memory is organised around peers (users, agents) and their sessions, rather than a flat chat log.

Managed or self-hosted

Use the managed api.honcho.dev service, or self-host from source with Docker Compose or a local dev setup.

Python and TypeScript SDKs

pip install honcho-ai, or npm install @honcho-ai/sdk, with equivalent workflows in both languages.

Agent-tool integrations

Documented integrations add Honcho memory to Claude Code, Codex and Cursor via dedicated packages and install scripts.

Built for stateful agents

Aims at continual, cross-session memory and context engineering rather than one-conversation chat history.

Advertisement
// selbst ausführen

Führe es selbst aus

  1. Clone the repository

    Get the source from the official GitHub repository for honcho.

    git clone https://github.com/plastic-labs/honcho.git
    cd honcho
  2. Install the SDK

    Install the Python SDK as the README quickstart shows (TypeScript: npm install @honcho-ai/sdk).

    pip install honcho-ai
  3. Or self-host from source

    Clone the repo and start the full service locally with Docker Compose, then set HONCHO_URL or base_url to your instance.

    git clone https://github.com/plastic-labs/honcho.git
    cd honcho
    docker compose up  # see README self-hosting section
// anwendungsfälle

Was du damit bauen kannst

Advertisement
// häufige fragen

Häufige Fragen

Who maintains honcho?

plastic-labs (Plastic Labs), through the plastic-labs/honcho repository, with docs at honcho.dev/docs.

Is honcho free to use?

Yes — the project is open source under the AGPL-3.0 license as recorded for this listing. See the license terms for any conditions on redistribution.

Does honcho need a paid API key?

The managed service uses api.honcho.dev (account/API key). Self-hosting is free under AGPL-3.0 and runs on your own infrastructure; reasoning uses models you configure there.

How does Honcho compare to storing chat history in my own database?

A database stores raw history; you still build recall, summarisation and context assembly. Honcho adds the reasoning layer — it processes messages in the background so queries return understanding and prompt-ready context, not just logs.

How do I get started with honcho?

Clone the repository (or use the documented installer), follow the install and configure steps on this page, and run the smallest real task first — store a few messages on a session, let background reasoning run, then query Honcho for what it knows about that peer and inject the context into a prompt.

// weiter entdecken

Ähnliche Tools

KI-Agenten

hindsight

Agent memory that learns — so your agents get smarter over time.

★ 47,0 Tsd. SterneMIT
Anleitung ansehen →
KI-Agenten

letta

Stateful AI agents with memory that learns and improves.

★ 25,1 Tsd. SterneApache-2.0
Anleitung ansehen →
KI-Agenten

mex

Team memory for engineers and their AI agents.

★ 1,8 Tsd. SterneMIT
Anleitung ansehen →
KI-Agenten

langmem

LangMem: long-term memory primitives for LangGraph agents that learn and adapt from conversations.

★ 1,7 Tsd. SterneMIT
Anleitung ansehen →
KI-Agenten

graphiti

Real-time knowledge graphs that give agents a memory with structure.

★ 31,5 Tsd. SterneApache-2.0
Anleitung ansehen →
KI-Agenten

Agent_Memory_Techniques

30 runnable notebooks teaching everything about agent memory.

★ 1,1 Tsd. SterneApache-2.0
Anleitung ansehen →
// alternativen

Alternativen im Bereich KI-Agenten

Die anderen Projekte der Kategorie KI-Agenten mit den meisten Sternen im LibreAI-Katalog — die realistische Auswahl, falls honcho nicht das Richtige ist.

// mehr in agents

Mehr in KI-Agenten

Verwandte Projekte der Kategorie KI-Agenten im Katalog, die honcho bei den Sternen am nächsten stehen — zum Weiterstöbern auf beiden Seiten.

Hol dir honcho

Kostenlos und offen — hol dir den Quellcode, gib dem Repository einen Stern und leg los.