Ask the model on my Mac
Ask a model served on the person's own Mac (a local model, free to run) a conversation or a yes-or-no question about some words, instead of a service billed per word
Runs on the person's Mac with the models it serves (computer/local-models.json: Ollama for conversations, a small classifier for yes-or-no questions). messages asks a conversation and answers in the chat-completions shape; tools optionally supplies the functions it may call, whose calls are returned for the caller to execute through its normal permissions. state and questions ask the classifier and answer its probabilities. Trillion asks it before any paid service for checking answers and for telling chatter from requests (src/model-routes.json). An explicitly chosen free always-on AI uses this same local tool. Answers error in words when no model on the Mac is running.
What it asks for
- job (Optional, words): A model job row from computer/local-models.json; the default row is used when absent
- messages (Optional, list): A conversation: [{ role, content }]
- tools (Optional, list): OpenAI-style function definitions the local model may request; the caller executes and checks them.
- max_tokens (Optional, number):
- temperature (Optional, number):
- seed (Optional, number):
- state (Optional, words): The words a yes-or-no question is about
- questions (Optional, details): The classifier's questions by id: { verdict: { type: 'noul', instructions, criteria } }
Use it from your AI
Connect the AI you already use to Trillion once. Then ask in your own words; your AI finds this tool by what it does and runs it.
Say to your AI: “Use Trillion to ask a model served on the person's own Mac (a local model, free to run) a conversation or a yes-or-no question about some words, instead of a service billed per word”
What your AI calls: run_tool {"tool_id":"tool:ask-a-model-on-this-mac"}