Spaces:
Running
Idea: long-term memory for the agent across restarts
Hi, I look at OSW Studio and I like very much that it is an agentic website builder where you can use local models with Ollama or LM Studio, and also that conversation and project state can recover after a refresh. When the local engine itself restart, the agent still have to recompute the long coding prompt and the project context from the beginning. Galahad is long-term memory for the agent: it save and restore the KV cache byte-for-byte across restarts, so the agent skip recomputation and keep memory between sessions. In tests, restore 2.8 to 4.3x faster than recompute, and it can keep 50M tokens on disk. Would you be open to try this with the Ollama or LM Studio path, or want help integrating? https://github.com/corbenicai/galahad
Sietse
Heya, and thanks for the suggestion. I had a look, and I don't think there's anything to integrate on the OSWS side. The KV cache lives inside the inference engine, and OSWS only talks to engines and inference providers over their HTTP APIs, so save and restore has to happen there. From the README, Galahad looks to run in llama.cpp, vLLM and SGLang rather than Ollama or LM Studio. OSWS already has a llama.cpp provider and takes any OpenAI-compatible endpoint, so a Galahad-backed server should work with it as is. If you have one running, I'd be curious to hear how it goes.