Provider Setup

ProtoAgent stores provider configuration in ~/.protoagent/config.json by default. The Python core owns the config schema and returns a redacted view to the Rust UI.

Supported Providers

Provider id	Kind	Default base URL or behavior
ollama	local server	http://localhost:11434
lmstudio	OpenAI-compatible local server	http://localhost:1234/v1
llama.cpp-server	local server	http://localhost:8080
llama.cpp-local	local files	Scans common model folders for .gguf files.
openai-compatible	OpenAI-compatible server	http://localhost:1234/v1 unless overridden.
openai	API	Requires OPENAI_API_KEY or stored config key.
anthropic	API	Requires ANTHROPIC_API_KEY or stored config key.
gemini	API	Requires GEMINI_API_KEY or stored config key.
deepseek	API	Requires DEEPSEEK_API_KEY or stored config key.

Select A Model

cargo run --manifest-path cli/Cargo.toml -- model

The picker:

1. Calls protoagent_core.agent_engine.list_models(validate_api_keys=True).
2. Displays local server status, configured API providers, and model choices.
3. Prompts for missing API keys when needed.
4. Saves the active provider/model through set_active_model().

Inside the TUI, use:

/model

Store An API Key

cargo run --manifest-path cli/Cargo.toml -- key openai

Inside the TUI:

/key openai

Keys are stored in the config file unless an environment variable is present. The displayed config is always redacted by visible_config().

Context Window For Ollama

Ollama is currently the provider with app-managed context-window control.

/context window 16k
/context window auto

The selected value is sent as Ollama num_ctx and also recorded in ProtoLink's LLMModelProfile, so the Rust context meter and ProtoLink metrics use the same window.

Validation Cache

API key validation is intentionally cached for a short time. A successful key validation is reused for about ten minutes, while uncertain or failed validation can be retried quickly. Live model runs call remember_valid_provider() so the UI can reflect a recently successful provider/model pair.