LLM governance / policy control planePolicy Gateway

Migrate from Azure OpenAI to Policy Gateway

Move your Azure OpenAI workloads to Policy Gateway to gain full control over safety filters, quotas, and audit logs.

Azure OpenAI enforces strict content filters that you cannot fully control or disable. Policy Gateway allows you to replace those provider-imposed restrictions with your own policy-as-code.

By migrating to Policy Gateway and the abliterated model, you gain granular control over what is allowed, redacted, or refused.

Updated Jan 21, 2026Policy Gateway

Quick start

Base URL
Example request
curl https://api.abliteration.ai/policy/chat/completions \
  -H "Authorization: Bearer $POLICY_KEY" \
  -H "Content-Type: application/json" \
  -H "X-Policy-User: user-12345" \
  -H "X-Policy-Project: finance-assistant" \
  -d '{
    "model": "abliterated-model",
    "messages": [{"role": "user", "content": "Summarize Q4 usage trends."}],
    "policy_id": "finance-policy"
  }'

Service notes

  • Pricing model: Usage-based pricing (~$5 per 1M tokens) billed on total tokens (input + output). See the API pricing page for current plans.
  • Data retention: No prompt/output retention by default. Operational telemetry (token counts, timestamps, error codes) is retained for billing and reliability.
  • Compatibility: OpenAI-style /v1/chat/completions request and response format with a base URL switch.
  • Latency: Depends on model size, prompt length, and load. Streaming reduces time-to-first-token.
  • Throughput: Team plans include priority throughput. Actual throughput varies with demand.
  • Rate limits: Limits vary by plan and load. Handle 429s with backoff and respect any Retry-After header.

Why migrate?

#

Azure OpenAI provides safety filters, but they are additive—you cannot remove the baseline censorship.

  • Full Control: Define exactly what is allowed. No hidden prompt filters.
  • Predictable Refusals: Replace generic 'I cannot help' messages with custom reason codes.
  • Audit Trails: Log every decision to your own SIEM (Splunk, Datadog) without relying on Azure Monitor.

Enforce per-user and per-project quotas

#

Quotas are driven by policy_user and policy_project_id tags.

  • Tag every request with policy_user for chargeback and usage tracking.
  • Use per-project keys for app-level budgets and isolation.
  • Quota violations return a policy decision with reason codes.
Enforce per-user and per-project quotas
{
  "policy_id": "finance-policy",
  "org_controls": {
    "user_quotas": true,
    "project_keys": true,
    "user_quota": { "requests": 500, "tokens": 100000, "window": "daily" },
    "project_quota": { "requests": 20000, "tokens": 5000000, "window": "monthly" }
  }
}

Tag and export audit events

#

Audit logs include decision metadata plus the tags you send with each request.

Export logs to the SIEM or log platform your security team already uses.

  • policy_user, policy_project_id, and policy_target appear in every audit event.
  • Export destinations: Splunk HEC, Datadog Logs, Elastic, Amazon S3, Azure Monitor / Log Analytics.
  • Policy Gateway security & privacy explains what is stored.

Shadow mode and canary rollouts

#

Safely test your new policies before switching over production traffic.

  • Shadow mode logs decisions without enforcement.
  • Canary mode enforces on a sample before full rollout.
  • Auto-rollback protects against decision spikes.
Shadow mode and canary rollouts
{
  "rollout": {
    "shadow": { "enabled": true, "sample_percent": 20, "targets": ["finance-assistant"] },
    "canary": { "enabled": true, "sample_percent": 5, "targets": ["finance-assistant"] },
    "rollback_on_spike": true
  }
}

Migration checklist

#
  • Define your policy rules, quotas, and reason codes in Policy Gateway.
  • Create projects and scoped keys per app or tenant.
  • Update your base URL to point to https://api.abliteration.ai/policy/chat/completions.
  • Start in shadow mode to verify decision logic.

Common errors & fixes

  • 401 Unauthorized: Check that your API key is set and sent as a Bearer token.
  • 404 Not Found: Make sure the base URL ends with /v1 and you call /chat/completions.
  • 400 Bad Request: Verify the model id and that messages are an array of { role, content } objects.
  • 429 Rate limit: Back off and retry. Use the Retry-After header for pacing.

Related links