Platform guide

Manage enterprise agent workflows in Token Pyxis, from workspace setup and model routing to usage reconciliation.

Enterprise provisioning and departments

Once the enterprise service agreement is signed, authorized Token Pyxis staff provision the workspace. Authorized enterprise administrators create accounts in Members, and console access follows each user’s role and department.

The enterprise sets the billing and access boundary. Use departments to assign members, API keys and budgets to teams or projects.

Enterprise administrator
Manage members, departments and budgets, issue project API keys and review usage within your assigned scope.
Department member
Roles govern access to keys and usage. Keys issued by an authorized department account are assigned to its department.
Procurement and finance
Use an authorized enterprise account to review wallet and subscription records, and submit or withdraw eligible invoice requests.

Keys and permissions

  • An authorized enterprise administrator issues keys by environment and application in API Keys. Save the secret at issuance, then configure model access, budget, rate limits, expiry and allowed source IPs where applicable.
  • To rotate a key, create its replacement, test it against the target model and disable the old key. If a key is exposed, disable it immediately and review its request history.
  • Model requests use API keys. Management APIs use sign-in JWTs. Eligibility checks can use enterprise, department, plan and key model restrictions.
Compare credentials and protocols

Model routing

The gateway resolves the model and endpoint capability, then selects eligible channels according to enterprise group, plan and channel configuration. For the same key and model, it prefers the last successful channel when that channel remains eligible.

Eligible upstream failures can try another channel for the same model. To fall back to another model, configure it explicitly in your application and check permissions, capabilities and rates. Before resending a partially streamed response or a submitted job, check the original request status.

Check model capabilities and list rates

Rate limits, budgets and plans

Rate limits govern throughput and concurrency. RPM means requests per minute and TPM means tokens per minute. Enterprise and department budgets and API key quotas can also affect whether a request is admitted.

Check effective model rates in Models and Pricing, plan coverage in Subscription, and balances and transactions in Wallet. Eligible usage can consume plan allowance first. The signed enterprise agreement governs overage settlement. Fixed-rate endpoints may charge per image, second or call.

See token rates and a worked example

Usage and reconciliation

  1. Save the x-request-id response header, along with the time, environment and model ID.
  2. Find the entry in Request logs by request ID. Check status, token and cache usage, then compare the usage value before allowance with the wallet debit.
  3. Group and export usage by model or department in Usage report. Match the request ID to the wallet transaction reference and reconcile subscription debits separately.

Troubleshooting and compliance

For 401, check the endpoint’s credential. For 402, check wallet balance and allowance. For 429, back off using Retry-After when present. Share the request ID and time with your administrator, and keep API keys and complete payloads private.

Request logs and optional payload capture are managed separately. Payload capture is disabled by default. See the security and compliance pages for data handling and generated-content labeling.

Business inquiries
Scan with WeChat to add the business team on WeCom