Service availability
Supported services, account requirements, and current API limits.
Internalize is available at convergingthought.com/internalize/app. Sign in with an email code, purchase credits, and use the Playground or a scoped server-side API key. There is no free trial balance.
Current surfaces
| Surface | Hosted behavior |
|---|---|
| Marketing and documentation | Public, without an account |
| Email-code sign-in and project setup | Clerk-backed accounts |
| Project settings and API keys | Persistent project state and explicit key scopes |
| Request, memory, and usage views | Actual jobs, versions, reservations and settled usage |
| Billing | Live Polar checkout; prepaid project credits |
| Inference and internalization | Enabled for authenticated projects with sufficient prepaid credit |
| Public TypeScript package | Not published; use the HTTP API or OpenAI-compatible clients |
Running examples
Follow the quickstart after funding your project. New API keys have inference permission by default. Enable learning explicitly for keys that will call internalize; management access is separate.
An unfunded project receives 402 with insufficient_credits. A request that exceeds the project’s spending limit receives spend_limit_exceeded. Adding credit does not change key permissions or bypass the project limit.
Requests run as durable jobs. Keep the returned identifier and poll the same job after an interrupted connection; do not create another paid operation just because your browser stopped waiting. Operator recovery and service capacity limits can temporarily delay work. See polling and recovery.
Billing
Usage is prepaid, with no free credit or automatic top-up. Current rates are $12/M input, $30/M output, $2.50/M verified cached input, $36/M training, and $0.30/GB-month retained memory. Learning has no flat fee and requires an explicit budget.
Polar processes payments. A verified payment event credits the selected project; returning from checkout alone does not. See Payments.
Memory lifetime and validation
Initial checkpoint retention is 24 hours. The console and API expose the actual expiry; automatic renewal is not available. An expired memory returns an explicit error rather than silently falling back to the base model.
Each learning job reports validation results. A rejected candidate leaves the active version unchanged. Evaluate the knowledge and tasks your application depends on before activating an update; validation does not guarantee every answer.
Supported model and API limits
GLM 5.3 is the only model in the current API catalogue. There is no customer-selectable model field or automatic fallback model. Other models are planned, but no availability date is promised here.
The API exposes asynchronous results and OpenAI-compatible buffered SSE: the final answer is validated before delivery, rather than streamed token by token during generation. Cancellation, user-configured webhooks, adapter export, subject deletion and a public MCP endpoint are not exposed. Do not use private execution routes to approximate them. If a missing capability is essential, contact team@convergingthought.com with the intended workflow.