Spaces:
Runtime error
Idea: long-term memory for the agent across restarts
I went through your Code Act Agent and I like the real loop: the CodeActAgent Mistral 7B GGUF on llama.cpp, the thought / execute / solution tags, and a Python REPL so the agent can actually run the code it write. On a Space that sleep (or when the GGUF have to load again), those five interaction steps and the long problem-solving prompt start over. Galahad is a 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 our tests, restore 2.8 to 4.3x faster than recompute, and it can keep 50M tokens on disk. If you would be open to try it with the llama.cpp path, or want help integrating, I am glad to help โ https://github.com/corbenicai/galahad
Sietse