Skip to main content
TokenLab is multi-format: you can keep OpenAI-compatible clients, Anthropic-native Messages calls, Gemini-native REST calls, and media endpoints in their natural shapes. The safest migration is not to translate every workload into one universal format. Choose the route that owns the behavior your application needs.

​
Route Mapping

Existing workloadTokenLab base URLPrimary endpointMigration note
OpenAI Chat Completionshttps://api.tokenlab.sh/v1/chat/completionsSmallest change for OpenAI-compatible chat and function calling
OpenAI Responseshttps://api.tokenlab.sh/v1/responsesUse when your app depends on Responses-specific input, tools, or output handling
Anthropic SDKhttps://api.tokenlab.sh/v1/messagesDo not append /v1 to the SDK base URL
Gemini RESThttps://api.tokenlab.sh/v1beta/models/:model:generateContentKeep Gemini-native fields on the Gemini route
Media generationhttps://api.tokenlab.sh/v1/images, /videos, /music, /3dDiscover models with recommended_for and expect async polling where documented
Management and billinghttps://api.tokenlab.sh/v1/management/...Use management tokens for server-side usage and billing reconciliation

​
OpenAI-Compatible Migration

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-tokenlab-key",
    base_url="https://api.tokenlab.sh/v1",
)

response = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{"role": "user", "content": "Hello from TokenLab"}],
)
Keep your existing retry, timeout, and streaming code, but validate model IDs with GET /v1/models before production traffic. For image generation, send model explicitly and read the image guide because image models differ more than chat models.

​
Anthropic Migration

from anthropic import Anthropic

client = Anthropic(
    api_key="sk-your-tokenlab-key",
    base_url="https://api.tokenlab.sh",
)

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain TokenLab in one sentence."}],
)
Use /v1/messages for Claude-native tool use, thinking flows, and Anthropic message semantics. Do not translate Anthropic-only fields through Chat Completions unless you intentionally want an OpenAI-compatible behavior change.

​
Gemini Migration

curl "https://api.tokenlab.sh/v1beta/models/gemini-3.5-flash:generateContent" \
  -H "Authorization: Bearer sk-your-tokenlab-key" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"Hello"}]}]}'
Keep Gemini built-in tools, File API references, cached contents, function declarations, and native content parts on /v1beta when your app depends on Gemini-native behavior.

​
Media Migration

  1. Query GET /v1/models?recommended_for=image|video|music|3d.
  2. Read tokenlab.public_contract_summary in list responses and the full tokenlab.public_contract where available.
  3. Send an explicit model, especially for image endpoints.
  4. Store task_id, poll_url, endpoint, model, and your own job ID for async jobs.
  5. Reconcile costs through usage records and billing_transaction_id, not provider task IDs.
Media workloads need their own rollout plan because latency, retries, and final assets behave differently from chat completions.

​
Production Rollout Plan

PhaseGoalChecks
1. InventoryList endpoints, models, request fields, streaming/async behavior, and billing ownerNo hidden provider-only fields are assumed public
2. Single-route pilotMove one endpoint and one model familyResponse shape, cost, and logs match expectations
3. Shadow or sampleCompare selected outputs against the previous providerUser-visible quality and latency are acceptable
4. Gradual rolloutIncrease traffic by key, org, or feature flagWatch 4xx, 5xx, latency, balance, and duplicate async jobs
5. CleanupRemove old provider path only after stable usageRollback path and support playbook are documented

​
Migration Pitfalls

  • Do not put every model behind one OpenAI Chat Completions path if your app needs native Anthropic, Gemini, or Responses behavior.
  • Do not assume old image defaults. Send model explicitly.
  • Do not retry async create requests without checking whether a task was already created.
  • Do not expose provider routing metadata in your logs or UI.
  • Do not compare billing with provider task IDs. Use TokenLab usage records.

​
API Reference

TopicReference
Multi-Format APIMulti-Format API
OpenAI SDKOpenAI SDK
Anthropic SDKAnthropic SDK
Gemini NativeGemini Native API
Image GenerationImage Generation
Async Jobs & PollingAsync Jobs & Polling