Base URL
https://api.codimixa.com/v1 is the public API base used by the current console configuration.
CODING AGENT API QUICKSTART
A small quickstart for a serious integration.
Use the OpenAI-compatible surface your client already understands, then verify the exact model and feature support in your account before production traffic.
Built around a stable OpenAI-compatible request shape.
REQUEST LEDGERillustrative
route /v1$0.42
in prompt context18.4k
out model output2.1k
↺ cached context6.8k
latency windowyour data · your keys
01 / FIRST REQUEST
Use a familiar request shape
The example uses a placeholder model ID and an environment variable for the key. Confirm both values in the console before you run it.
shell
curl https://api.codimixa.com/v1/chat/completions \
+ -H "Authorization: Bearer $CODIMIXA_API_KEY" \
+ -H "Content-Type: application/json" \
+ -d '{"model":"YOUR_MODEL_ID","messages":[{"role":"user","content":"Review this function for a null handling bug."}]}'02 / BOUNDARY
Keep secrets and agent state in the right place
The website explains the public shape. Your client should hold the key, conversation state, repository context, and tool permissions. The gateway should receive only what the request requires.
Authentication
Send a bearer key from an environment variable. Never commit it to a repository or place it in browser code.
Model ID
Use the identifier shown for your account. A placeholder is safer than an unverified public model promise.
Response handling
Handle non-2xx responses, timeouts, and rate limits in the client before you add retries.
03 / BEFORE PRODUCTION
Make one small test set
Run a repeatable set of repository tasks, capture the output and token counts, and keep the model choice in configuration. That gives your team a baseline it can revisit after a route change.
- 01
Test a read
Ask the tool to explain a known function.
- 02
Test a patch
Give it a narrow bug with a clear expected outcome.
- 03
Test a loop
Measure a task that uses tools more than once.
NEED ACCESS?
Finish setup in the console.
The website does not create keys. Use the authenticated console for that boundary.
REQUEST ANATOMY
Know what each field is doing
The request is small, but each field has a job. Make the model ID and key configurable before you add retries or parallel work.
| Field | Purpose | Failure to handle |
|---|---|---|
| Authorization | Identifies the account key | Leaking the key or receiving 401 |
| model | Selects an account-visible route | Unknown or unavailable model |
| messages | Carries instructions and context | Wrong role or oversized context |
| Content-Type | Tells the gateway how to parse the body | 400 response or malformed payload |
ERRORS ARE PART OF THE API
Build the failure path before the happy path
A coding tool should be able to stop cleanly when the gateway rejects a request. Log a request identifier when the response provides one, but do not log the bearer key or full sensitive context.
401Authentication
Check the key source, environment, and revocation state.
400Request shape
Validate model ID, JSON, roles, and required fields.
429Rate limit
Back off with a bounded retry policy and visible stop condition.
5xxUpstream failure
Do not start an unbounded retry loop; preserve the task state.
PRODUCTION CHECK
A quick readiness review
Run one read-only task, one narrow patch, and one tool loop. Verify that the client can stop, retry once with a bound, and show a useful error to the operator.
Then move the key into the deployment secret store, keep the model ID in configuration, and record the exact base URL and route that were tested.