Discover and explore top open-source AI tools and projects—updated daily.
Kohaku-LabBuild and compose AI agents and multi-agent teams
Top 96.7% on SourcePulse
Summary
KohakuTerrarium offers a general-purpose AI agent framework and application for building and composing self-contained agents and multi-agent teams. It eliminates the need to repeatedly rebuild agent infrastructure by providing a reusable substrate, ideal for developers requiring flexible agent architectures, embedding agent behavior in Python, or managing evolving requirements.
How It Works
The core abstraction is the "creature," a standalone agent with modules for control, tools, triggers, memory, and I/O. Creatures compose horizontally into a "terrarium," a wiring layer managing lifecycles and communication. This design prioritizes configuration and module customization over reimplementing agent plumbing, accelerating development. Its Python-native nature enables deep integration.
Quick Start & Requirements
Install via pip: pip install kohakuterrarium. Fetch OOTB agents with kt install https://github.com/Kohaku-Lab/kt-biome.git. Authenticate LLM providers (OpenAI, Anthropic, Gemini, OpenRouter) using kt login codex. Run agents from CLI (kt run @kt-biome/creatures/swe --mode cli), manage multi-agent systems (kt terrarium run @kt-biome/terrariums/swe_team), or launch a web dashboard (kt serve start). Documentation and tutorials are readily available.
Highlighted Details
kt resume.>>, &, |.kt install.Maintenance & Community
Maintained by Shih-Ying Yeh (KohakuBlueLeaf) and contributors. Community channels include QQ (1097666427), Discord (https://discord.gg/xWYrkyvJ2s), and a forum (https://linux.do/). A roadmap is available.
Licensing & Compatibility
Released under "KohakuTerrarium License 1.0," based on Apache-2.0 with naming and attribution requirements for derivative works. Derivative projects must include "Kohaku" or "Terrarium" in their name and provide visible attribution with a link, potentially impacting commercial use.
Limitations & Caveats
Not recommended for use cases requiring sub-50ms per-operation latency. The "terrarium" architecture for multi-agent systems is noted as still evolving.
18 hours ago
Inactive
nicobailon
letta-ai