Build on the models
behind Saanora.
One API, one key and one prepaid balance for every Saanora model. Start with Clara, a decision model that matches Jev on tool routing and the best open model on coherence. Mark, the model behind Saanora Chat, joins next.
- latency
- 0 ms
- tokens
- 316
- decisions
- 4
- cost
- $0.000158
Our nightly export has failed three nights in a row. Finance closes the quarter tomorrow and needs the file by 9am.
→ If you can ask it, Clara can decide it.
Routing, triage, risk, moderation, compliance, agent control. Write the question and the answers you accept; Clara returns a calibrated probability for each one, for any record, in any domain.
Jev-level decisions.
Built by Saanora.
On AgentDecisionBench Clara lands within 0.05 points of Jev 1.13 and ahead of the OpenAI Decisions API. On JevBench-mini it is level with the leaderboard's #1 and leads it on three of five relation families. Measured on public benchmarks Clara never trained on; third-party rows are each benchmark's published results. Where Clara is weaker.
Within 0.05 points of Jev 1.13 and ahead of the OpenAI Decisions API, on a benchmark Clara never trained on. Axis from 85%.
Point it at Saanora with a Saanora API key. Same request body, same answers shape, Clara on the other end.
- base_url = "https://your-current-decision-api"+ base_url = "https://api.saanora.com"+ api_key = "sk_sa_..." # sent as Authorization: Bearer # POST /v1/systemone · same body, same answers shape
Clara speaks the open System One format, so harnesses such as JevBench run against it unchanged.
One platform for every Saanora model.
Each model is one field in the request. Keys, limits, credit and usage stay the same as the catalog grows.
clara-1Level with Jev on tool routing and with the best open model on coherence. Yes/no, choice and score answers with calibrated probabilities, 64 questions about one record per call.
mark-1The model people already talk to in Saanora Chat: writing, research, documents and code. Arriving on the same keys, credit and limits, priced per input and output token.
One request.
Every decision about the record.
No prompts to tune and no text to parse. Describe the record once, ask everything you need to know, and branch on the probabilities.
A ticket, transaction, agent trace, invoice or any document, as text or JSON. Up to 16K tokens.
Our nightly export has failed three nights in a row. Finance closes the quarter tomorrow.
Yes/no, one of up to 255 options, or a point on a scale. Up to 64 questions about the same record, in one call.
- noulfailing · Is an export failing?
- choiceteam · Which team owns it?
- scoreseverity · How severe is it?
Every option comes back with a probability, and they sum to 1. Your thresholds live in code you can test and review.
The decisions your business makes ten thousand times a day.
Too many for people, too important for guesswork. Each of these is one request with a handful of questions.
[0] find_customer({email}) → C-2207, gold [1] list_orders(C-2207) → A-1043 shipped, A-1051 pending Tools: get_order · cancel_order · issue_refund · ask_user · finish
next_toolchoiceneeds_humannoulFeels like the APIs you already use.
Plain JSON over HTTPS with the same conventions on every model: request ids, typed errors, request and token rate-limit headers, and a System One endpoint for existing harnesses.
curl --retry 5 https://api.saanora.com/v1/decisions \
-H "Authorization: Bearer $SAANORA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "clara-1",
"state": "Customer writes: our nightly export has failed three nights in a row and the finance team needs it by 9am.",
"questions": {
"failing": { "type": "noul", "instructions": "Is a scheduled export currently failing?" },
"team": { "type": "choice", "instructions": "Which team should own this?",
"criteria": { "engineering": "Fixes broken product functionality",
"billing": "Handles invoices, charges and refunds" } },
"severity": { "type": "score", "instructions": "How severe is the business impact?",
"criteria": ["none", "low", "high", "critical"] }
}
}'{
"id": "dec_1eab8757a66cc4fba1e7d9f7",
"object": "decision",
"created": 1792431238,
"model": "clara-1",
"mode": "fast",
"answers": {
"failing": { "type": "noul", "noul": 0.996 },
"team": {
"type": "choice", "choice": "billing", "confidence": 0.317,
"probabilities": { "engineering": 0.341, "billing": 0.659 }
},
"severity": {
"type": "score", "score": 2.48, "confidence": 0.511,
"probabilities": { "0": 0.001, "1": 0.004, "2": 0.511, "3": 0.484 },
"legend": { "0": "none", "1": "low", "2": "high", "3": "critical" }
}
},
"usage": { "input_tokens": 191, "decisions": 3, "cost_usd": 0.000096 },
"latency_ms": 279.6
}HTTPS under /v1, a model field on every request, OpenAI-style list objects, ids and typed error bodies.
Authorization: Bearer sk_sa_… (or X-API-Key). Stored only as hashes; revoked keys stop within 60 s.
Yes/no returns p(yes); choice and score return a probability for every option, summing to 1.
POST /v1/systemone takes the open System One wire format, so existing clients switch by base URL.
Everything a team needs to ship it.
Organizations, projects, spend caps and an audit trail of every request, built in from the first key.
Invite people as owner, admin, developer or billing. Access is enforced in the database, not just the UI.
Separate keys and usage per app or environment, with a monthly limit that stops spend at the line.
Top up from $5; credit never expires. Each request is priced up front and deducted exactly, so the bill never surprises you.
Request id, latency, tokens and cost for every call, by model, project and key. Export it as CSV. Failed calls are free.
Up to 5,000 requests and 5M tokens a minute, unlocked automatically as your organization grows. Dedicated capacity beyond that.
We meter counts, tokens and cost. The records and questions you send are processed in memory and never written to our database.
Know the bill before you ship.
Pay per token, like every major model API. Clara's output is $0; you pay only for the input it reads. No seats, no minimums. See pricing.
Make your first decision today.
Create an account, add credit, get a key and send your first request in under five minutes.