Which API should you use with Docker Agent?

For Docker Agent, Docker's open-source runtime for YAML-defined agents, the best API for agents that run every day is a cost-optimized OpenAI-compatible provider: define a provider with base_url https://api.stdcmpt.com/v1, token_key STANDARDCOMPUTE_API_KEY and api_type openai_chatcompletions, then use standardcompute/standardcompute as the model, and Standard Compute's router sends each step to the cheapest model that can do it (Claude Fable 5.1 or GPT-6 Astra only when it counts), on a plan from $19 a month with a $20 compute budget. Docker Agent's built-in providers (Anthropic, OpenAI, OpenRouter and more) bill per token, and Docker Model Runner runs open models locally for free. Docker Agent itself is free (Apache 2.0).

Light or occasional use

For experiments or small agents, Docker Model Runner costs nothing per token if your machine can run a capable open model. For light hosted use, a pay-per-token key on an efficient model will likely cost the least.

Daily use, cost-optimized

When your agent teams run every day with sub-agents and tool calls, a Standard Compute provider sends each step to the cheapest model that can do it, and for every model we vet providers for price, uptime and speed, with automatic fallback. Opt-in Europe or US Compliance rules apply per key, team or company, and we don't train on your prompts or store them. The bill stays at the monthly plan price: requests stop with HTTP 402 when the compute budget is used, with nothing billed on top.

API and subscription options for Docker Agent

Prices and allowances can change. Docker Agent's provider fields, global providers, secret lookup and Docker Model Runner details come from its provider-definition, secrets and model-runner docs (main-branch docs, global providers checked in the v1.150.0 release), read on October 15, 2026. Docker Agent: provider definitions · Docker Agent: managing secrets · Docker Agent repo. Spot an error? Tell us.

Add Standard Compute as a Docker Agent provider

providers:
  standardcompute:
    base_url: https://api.stdcmpt.com/v1
    token_key: STANDARDCOMPUTE_API_KEY
    api_type: openai_chatcompletions

agents:
  root:
    model: standardcompute/standardcompute
    description: A helpful assistant
    instruction: You are a helpful assistant.
    sub_agents: [reviewer]
  reviewer:
    model: standardcompute/anthropic/claude-sonnet-5.5
    description: Reviews changes
    instruction: You review code.

Put STANDARDCOMPUTE_API_KEY in your environment or ~/.config/cagent/.env, then run docker agent run agent.yaml. To register the provider once for every command, run docker agent setup and pick Custom OpenAI-compatible endpoint; then docker agent run --model standardcompute/standardcompute works anywhere.

Docker Agent questions

Why does Docker Agent send my custom model to the Responses API?

When api_type is not set, Docker Agent picks the API from the model name: newer OpenAI-style names (gpt-5, o-series, Codex) default to openai_responses and everything else to openai_chatcompletions. Set api_type: openai_chatcompletions on the provider to make it explicit. Standard Compute serves both at https://api.stdcmpt.com/v1.

Why does model: auto never pick my custom provider in Docker Agent?

Automatic model selection skips custom providers by design. Reference the model explicitly, for example model: standardcompute/standardcompute in the agent file, --model standardcompute/standardcompute on the command line, or default_model.

Where does Docker Agent read the key named in token_key?

From a chain of secret sources, first match wins: environment variables, Docker Compose secrets in /run/secrets, the env file ~/.config/cagent/.env (written by docker agent setup and read on every run), a credential helper and Docker Desktop's secret store. You can also pass a file with --env-from-file. Current versions no longer read the macOS Keychain or pass, so re-run docker agent setup if your key lived there.