Llama API (Meta)
Meta's official Llama API. Bring your own key; nothing is billed by us.
What the harness already knows
| Provider id | llama |
| Base URL | https://api.llama.com/compat/v1 |
| Protocol | OpenAI-compatible chat completions |
| API key | LLAMA_API_KEY |
| Model discovery | No — models come from the catalogue and from what you configure. |
Connecting it
The endpoint above is built in, so there is nothing to configure. Put the key in the environment and the provider appears in the model picker:
export LLAMA_API_KEY="..."Keys are issued by Llama API (Meta) — where to get one.
Models on Llama API (Meta)
The newest models Llama API (Meta) serves that can call tools, from the models.dev snapshot OHarness ships.
| Model id | Context / output | $/M in / cached / out | Reasoning |
|---|---|---|---|
cerebras-llama-4-maverick-17b-128e-instruct | 128K / 4K | 0 / — / 0 | No |
cerebras-llama-4-scout-17b-16e-instruct | 128K / 4K | 0 / — / 0 | No |
groq-llama-4-maverick-17b-128e-instruct | 128K / 4K | 0 / — / 0 | No |
llama-4-maverick-17b-128e-instruct-fp8 | 128K / 4K | 0 / — / 0 | No |
llama-4-scout-17b-16e-instruct-fp8 | 128K / 4K | 0 / — / 0 | No |
llama-3.3-70b-instruct | 128K / 4K | 0 / — / 0 | No |
llama-3.3-8b-instruct | 128K / 4K | 0 / — / 0 | No |
List prices in USD per million tokens, read from models.dev on 2026-09-19. Llama API (Meta) changes its prices and models; its own pricing page is the authority. Start a session on the newest:
oharness -m llama/cerebras-llama-4-maverick-17b-128e-instructPointing it somewhere else
A regional host, a gateway, or a server of your own. Overrides in ~/.oharness/config.json are field level, so changing the endpoint does not mean restating the rest of the provider — this is the whole block, shown holding the default so the shape is clear:
{
"providers": {
"llama": {
"baseUrl": "https://api.llama.com/compat/v1"
}
}
}What you get either way
The same agent loop, the same tools, the same MCP servers and the same permission rules, whichever provider is underneath — and switching between them mid-session converts nothing, because history is never stored in a vendor’s shape. What an agent harness is covers why that is the hard part, and the providers reference covers the catalogue itself.