diff --git a/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml b/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml index 28681d4cc..db38e82d4 100644 --- a/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml +++ b/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml @@ -4,8 +4,16 @@ cloud_name: Refact local self-hosting server #chat_endpoint: "http://localhost:8008/v1/chat/completions" #chat_model: "qwen2.5/coder/1.5b/instruct" +# Example of self-hosting via LM Studio +chat_endpoint: "http://localhost:1234/v1/chat/completions" +chat_model: o3-mini +# Be sure to load any model you prefer in LM Studio first, it will not check the name of a model set there, +# but using "known" for Refact model name like "o3-mini" will allow use not only "Quick", but also "Explore" and "Agent" modes. +# for gpt-4o deep_think tool does not work, for o3-mini works + embedding_endpoint: "http://localhost:8008/v1/embeddings" -embedding_model: "thenlper/gte-base" +embedding_model: "thenlper/gte-base" # or "thenlper/gte-base/cpu" +embedding_size: 768 # mandatory to be set completion_endpoint: "http://localhost:8008/v1/completions" completion_model: "Refact/1.6B"