Open weights, hosted for you
DeepInfra hosts open-weight models behind an API, so agents run them without you running hardware.
DeepInfra
Hosted open-weight models behind one key — the lane for agents that think on open models.
model lane · your key, DeepInfra's prices
Missions
What agents get done with it.
Priced per token, by them
DeepInfra meters the key you connect per token at its own prices; Ahel adds nothing on top.
The picker follows the key
Once connected, the model picker fills with the DeepInfra models your key can reach; you pick per agent.
The catalog
Every action, honestly listed.
A model provider gives an agent a lane to think on, not actions to call — so there is no action catalogue here. Tool actions come from the other integrations.
Connect
Your key, their prices.
- Your key
- pasted once in your Ahel workspace · encrypted at rest, never shown back to you · removable any time
- Markup
- none — DeepInfra usage at the vendor’s own prices
- Auth
- your own DeepInfra key, from DeepInfra's dashboard
- Billing
- per-token, billed by DeepInfra — no markup
- Models
- hosted open weights, not one vendor's frontier line
- Scope
- thinking only — no tool actions ride this lane
Your own DeepInfra key, created in their dashboard and pasted once into your Ahel workspace.
FAQ
Questions people actually ask.
Where do I get a DeepInfra API key?
From DeepInfra's own dashboard; paste it once in your Ahel workspace and it is stored encrypted. Ahel is invite-only right now, with access opened in batches.
What does DeepInfra cost through Ahel?
DeepInfra bills your key per token at its own prices, which vary by model; Ahel adds no markup. DeepInfra's model pages list the current rates.
Why run open-weight models at all?
Some operators want models whose weights are public, hosted by someone else — DeepInfra is that hosting. Whether a given model suits a job is your call; the picker offers what your key reaches.
OpenRouter
Venice AI