Tase Platform
Build on the Tase Platform.
The models and infrastructure behind the Tase assistant, exposed as an API. Chat, tool calling, streaming, and stored agents — with a console for keys, usage, and cost.
$5 of credit on sign-up · no card required
curl https://tase.app/api/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TASE_API_KEY" \
-d '{
"model": "tase-0.1-8b",
"messages": [
{ "role": "user", "content": "Log a $12 lunch and remind me to invoice Acme." }
]
}'Create a key in the console, export it as TASE_API_KEY, paste, run.
API surface
Everything the assistant runs on, exposed.
The model
One model, trained for the work.
tase-0.1-8b
Tool calling, low latency, Turkish + English
The default model. Specialised for picking the right tool, producing correct arguments, and answering fast. Tool schemas are trained into the weights, so requests stay small. Long-horizon requests escalate automatically — the API surface does not change.
- Context
- 128K
- Max output
- 8K
- Input
- $0.15/ 1M
- Output
- $0.60/ 1M
Console
The parts of an API you live with.
Keys, a workbench, usage, cost, request logs, rate limits, and prepaid credit — all in one place, all reading the same data the API writes.
Keys you can actually manage
Create, name, and revoke per service. Per-key request counts, tokens, and spend, so you know what a key is doing before you rotate it.
Workbench on the real endpoint
Test prompts against the same path, the same model, and the same meter as production — not a sandbox with different latency.
Usage, cost, and logs
Request-level logs with tokens, cost, and latency. Daily rollups by model, key, and surface. Failed requests are logged and never charged.
Limits without a sales call
Rate limits are published, enforced per key, and rise automatically as your credit balance grows.
Plans
Prepaid credit. Published limits.
Free
Build and test. $5 of starter credit, no card required.
- ·$5 starter credit
- ·20 requests / minute per key
- ·2 concurrent requests
- ·Full API surface — chat, streaming, tools, agents
Scale
Production throughput. Unlocked automatically at $20 in credits.
- ·120 requests / minute per key
- ·12 concurrent requests
- ·20 API keys, 50 stored agents
- ·Usage + cost breakdown by key
Enterprise
Custom limits, invoicing, and a named contact.
- ·Negotiated rate limits and concurrency
- ·Invoiced billing instead of prepaid credits
- ·Named technical contact
- ·Priority incident response
What people build with it.
Turn messages into actions
Take a line of natural language — "log a $12 lunch", "move my 3pm to Thursday" — and get back a tool call your backend can execute.
Classify at volume
Receipts into categories, tickets into queues, notes into projects. Small model, low latency, priced for the volume classification actually runs at.
Assistants that remember
Feed your own context in, get structured output back, and store what matters. The API is stateless so your memory layer stays yours.
Voice and chat front-ends
Stream tokens into a UI as they generate. The event format is the one your existing chat client already parses.
Elite Mentorship
Want the person who built this in your repo?
A small number of seats each cohort: weekly 1:1s, hands-on architecture and code review, and a launch plan that covers distribution, not just the build. Application-gated, because a seat is real calendar time.
Questions
Is the API really OpenAI-compatible?+
The chat-completions request and response shapes match, including streaming frames and tool calls. In practice you change two things: the base URL and the key. Anything genuinely Tase-specific — agents, the usage endpoint — lives on its own paths so it cannot collide with a client library update.
What does it cost to start?+
Every account gets $5 of credit on first sign-in, with no card. That is enough for tens of millions of input tokens at list price. Beyond that you buy credit in packs; nothing auto-renews, and requests stop rather than generating an overage bill.
Do I need a Tase subscription?+
No. Platform credit and the consumer Tase subscription are separate. An API customer does not need a Pro or Max plan, and a Pro subscriber does not automatically get API throughput.
How do rate limits work?+
Limits are per API key, published per plan, and returned on every response as x-ratelimit-* headers. Over the limit you get a 429 with Retry-After. The Scale tier unlocks automatically once your lifetime top-ups reach $20 — no request, no sales call.
Where does my data go?+
Requests are served, logged for 90 days with metadata (model, token counts, latency, status), and billed. Request logs record the shape of the call, not the content of your prompts. Agent runs store the input and output you send, because showing you a run history is the point of them.
What happens if a request fails?+
You get a typed error envelope and no charge. Failures still appear in your logs so you can see the pattern, but a 502 on our side never touches your balance.
First call in five minutes.
Create a key, paste the snippet, watch the tokens land in your usage page.