Connect the OpenAI API¶
Goal¶
Route the primary Colossus model through the public OpenAI Responses API while keeping the API key outside configuration.
Prerequisites¶
- OpenAI API access, API billing, and an API key available to the Colossus process.
- A schema version 2 configuration generated by
colossus config init. - An exact model identifier plus its context, output, tool-call, and streaming limits.
OpenAI API access is separate from a ChatGPT subscription. If you intend to use an eligible ChatGPT/Codex plan instead of API billing, follow Codex or ChatGPT subscription.
Steps¶
1. Inject the API credential¶
Set OPENAI_API_KEY with your platform's process environment or secret manager before
starting Colossus. The YAML below contains only env:OPENAI_API_KEY; do not replace that
reference with the secret value.
2. Configure the Responses route¶
Run colossus -w . config effective, edit the reported resolution.configPath, and
keep its required storage block and any intended custom settings. Apply this validated
overlay, then replace
YOUR_OPENAI_MODEL_ID and the model limits with exact values for the selected model.
schemaVersion: 2
providers:
profiles:
openai-provider:
kind: open_ai_responses
baseUrl: https://api.openai.com/v1
credentialReference: env:OPENAI_API_KEY
models:
profiles:
openai:
providerProfile: openai-provider
model: YOUR_OPENAI_MODEL_ID
contextWindowTokens: 128000
maxOutputTokens: 16000
capabilities:
toolCalls: true
streaming: true
roles:
primary: openai
sandbox:
networkDestinations:
- https://api.openai.com
Colossus appends /responses and /models to the API prefix. Keep /v1 in baseUrl.
The shown exact https://api.openai.com sandbox grant applies under an isolating
boundary; acknowledged full access needs no duplicate grant, and adding one does not
narrow ambient authority. Set model limits
and capabilities from the selected model's documentation; Colossus does not infer them
from the catalog.
3. Inspect routing and readiness¶
colossus -w . models route primary
colossus -w . provider doctor openai-provider
colossus -w . provider models openai-provider
colossus -w . models doctor openai
4. Send one bounded model turn¶
Expected result¶
The primary role resolves to openai, provider and model diagnostics succeed, and the
run returns connected.
Verification¶
Run colossus -w . config show and confirm that the output contains the credential
reference but not the API key value. Under isolation, also confirm the exact sandbox
origin; under full access, confirm config effective reports ambient network authority.
Failure path¶
- Credential unavailable or HTTP 401: confirm
OPENAI_API_KEYis present in the environment of the Colossus process, then rerunprovider doctor openai-provider. - The model is absent or denied: run
provider models openai-provider, copy an exact accessible model ID, and rerunmodels doctor openai. - The provider origin is denied under isolation: grant exactly
https://api.openai.cominsandbox.networkDestinations; do not put/v1there. - Generation rejects the request: verify the selected model's limits and set
toolCallsorstreamingtofalsewhen that model does not support the contract, then rerunmodels doctor openai.
Next step¶
Review exact field semantics in Providers and models, apply deployment guidance from Providers and routing, or continue with First repository task.