HumboldtDocs
Use Hum

Models and credits

Routed models

Signing in gives you a key on Metaphi AI's model proxy.

/model            # list, cheapest first; a number switches
/model <name>
/cost             # tokens and dollars this session

Credits

/usage-credits            # balance and the page that adds more
hum credits <code>        # redeem a code

When credits run out, Hum opens the credits page.

Your own host

Any OpenAI-compatible endpoint. Sessions on it call the host directly; every model call is kept under ~/.hum/calls/.

[model]
name = "my-model"
api_base = "https://llm.example.com/v1"
api_key_env = "MY_KEY"
context_window = 200000
hum whoami        # signed in? whose model?

Sub-agents

[model]
agents = "deepseek-v4-flash"     # a cheaper route for sub-agents