Cognocient troubleshooting — common issues and fixes

This page covers the most common issues after setup, with step-by-step fixes. If you don't find your answer here, email hello@cognocient.com — we respond within one business day.

Calls not appearing in the dashboard

You've integrated the proxy and your app is running, but the Calls page shows no data.

Check 1 — Is base_url actually pointing to Cognocient?

The most common cause: your OpenAI client is still pointing to api.openai.com. The proxy only sees calls that come through it.

# ✗ Wrong — calls go to OpenAI directly, not logged
client = OpenAI(api_key="sk-cog-YOUR-KEY")
 
# ✓ Correct — both changes are required
client = OpenAI(
    api_key="sk-cog-YOUR-KEY",
    base_url="https://api.cognocient.com/v1"
)

Check 2 — Are calls actually reaching the proxy?

Run a single test call with curl. If it returns a valid response, the proxy is working and calls will appear in the dashboard within a few seconds.

curl https://api.cognocient.com/v1/chat/completions \
  -H "Authorization: Bearer sk-cog-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -H "X-Cost-Feature: test" \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"hi"}]}'

Check 3 — Feature header missing

Calls without an X-Cost-Feature header are attributed to untagged. They still appear — filter by feature "untagged" in Live Calls.

Getting 401 errors from the proxy

The proxy returns 401 when it cannot authenticate your request. This is different from a 401 from the AI provider.

Cause: Wrong key type

You are passing your OpenAI or Anthropic key directly, not a Cognocient proxy key. Proxy keys always start with sk-cog-.

Fix: Generate a key in Dashboard → API Keys. Use that as your api_key — do not use your provider key.

Cause: Key was revoked

API keys can be revoked from the dashboard. If a key is deleted, all requests using it immediately return 401.

Fix: Dashboard → API Keys → confirm the key is still listed and active.

Cause: Trial expired

The 10-day free trial has ended and no payment method has been added.

Fix: Dashboard → Billing → add a payment method. Calls resume immediately.

Budget not blocking calls

Cause 1: Enforcement mode is "Alert", not "Block"

Fix: Go to Dashboard → Budgets → edit the budget → change mode to Block or Degrade.

Cause 2: X-Cost-Feature header doesn't match the budget's feature name (case-sensitive)

Budget feature names are case-sensitive. "Chatbot" and "chatbot" are different features. Confirm the exact string in your header matches the budget configuration.

Cause 3: Spend hasn't exceeded the limit yet

Block mode triggers when spend exceeds the limit. Alert mode triggers at the threshold (default 90%). Check current spend vs. limit in Dashboard → Budgets.

Wrong model being used / model not found

Your app requests gpt-4o but the response comes from gpt-4o-mini, or you get a 404 "model not found" error.

Model was degraded by a budget rule

Check the response headers. If x-cog-degraded: true is present, a budget in Degrade mode downgraded the request.

Fix: either increase the budget threshold, switch to Block mode, or accept Degrade as intended behaviour.

Model name not recognised by provider

The proxy forwards the model name exactly as you send it. Use the exact model ID from the provider's documentation (e.g. gpt-4o, not gpt4o). Check Supported Providers for the model IDs we support.

PDF report not generating

You click "Generate AI Report" and nothing downloads, or you see an error.

Cause 1: No spend data for the selected period

Fix: Reports require at least one call in the selected month. Run a test call through the proxy first, then try generating the report.

Cause 2: Report generation timed out

Fix: The AI narrative takes 15–20 seconds. If you get a timeout, wait a moment and try again — the backend may be under load.

Cause 3: No outcomes tracked (Efficiency Score missing)

Fix: The AI Efficiency Score requires at least one outcome tracked via Dashboard → Outcomes. The report still generates without it, but the efficiency section will be omitted.

"sk-cog-..." is not a valid API key

OpenAI or Anthropic SDKs reject your proxy key with a "not a valid API key" error.

Both changes are always required together:

# ✗ Wrong — base_url defaults to api.openai.com
client = OpenAI(api_key="sk-cog-YOUR-KEY")
 
# ✓ Correct — both point to Cognocient
client = OpenAI(
    api_key="sk-cog-YOUR-KEY",
    base_url="https://api.cognocient.com/v1"
)

// ✓ Node.js
const openai = new OpenAI({
  apiKey: 'sk-cog-YOUR-KEY',
  baseURL: 'https://api.cognocient.com/v1',
});

Next steps: Error Reference · Quickstart · Contact Support