Honcho
Honcho is an AI-native memory platform developed by Plastic Labs that provides stateful, continually learning context for AI agents. Rather than simply storing and retrieving facts, Honcho uses custom reasoning models powered by Neuromancer to analyze every message, infer conclusions, detect patterns across interactions, and generate hypotheses about users. This reasoning layer is designed to deliver the context that matters while reducing token consumption, with claimed token savings of 60-90% compared to naive retrieval approaches. The platform exposes a simple API: developers write messages to Honcho, which are stored and indexed, while Neuromancer automatically reasons over them. The context() method returns instantly curated reasoning plus conversation history, giving agents continuity at every turn. For more complex queries, chat() provides on-demand advanced reasoning with configurable resource tiers ranging from Minimal to Max, allowing developers to balance cost, latency, and depth of analysis. Honcho is model-agnostic and works with any stack or scale. It offers integrations with popular agent frameworks and coding tools, including Claude Code, OpenAI Codex, OpenClaw, Hermes Agent, and DeepSeek Harness. The platform also provides a CLI for quick setup and agent skills for developers. Honcho reports state-of-the-art results on benchmarks such as LongMem, LoCoMo, and BEAM, and is available as an open source project on GitHub under plastic-labs/honcho, with a hosted app at app.honcho.dev.
- Category
- AI Agent
- Pricing
- Unknown
Problem solved
Traditional memory systems store facts but fail to reason about them, wasting tokens and missing important context. Honcho solves this by continually learning from user interactions, automatically reasoning over stored data, and providing agents with instant, curated context that improves statefulness and reduces token usage.
Use cases
- Building stateful chatbots that remember user preferences and history
- Providing persistent memory for AI coding assistants like Claude Code and Codex
- Adding long-term memory to autonomous agents and agent frameworks
- Managing context for customer support agents with user interaction history
- Creating personal AI assistants that learn across sessions
- Enabling messaging platform bots (WhatsApp, Telegram, Discord, Slack) with memory
- Reducing token costs in LLM applications by delivering only relevant context
Audience
- Developers
- ML Engineers
- Data Scientists
- Researchers
- Startups
Pros
- State-of-the-art benchmark results on LongMem, LoCoMo, and BEAM
- Significant token savings through automatic reasoning and context curation
- Model-agnostic and works with any LLM or stack
- Easy integrations with popular agent tools and frameworks
- Open source with a hosted cloud option
Cons
- Pricing is not publicly listed
- Requires API key and integration setup
- Reasoning layer may introduce latency (~200ms per ingestion)
- Documentation may be complex for new users
- Project is relatively new and evolving rapidly
Official website: https://honcho.dev/
Canonical LambdaBase page: https://www.lambdabase.com/tools/honcho