Typed decisions with probabilities in milliseconds.
Classify incoming packets and branch your code on a boolean, label, score or text span with a measured probability. Clear cases clear Decide at 600 ms p90; ambiguous ones escalate to a larger model or a reviewer.
- w9-form.pdf tax_form 0.98 Straight through
- march-statement.pdf bank_statement 0.95 Straight through
- scan-0042.jpg claim_letter 0.61 To review
threshold 0.90
Classify: Category labels with confidence scores
Send any PDF, scan, office file or audio recording with 2 to 50 classes described in plain language, with no training data required. Classify whole documents quickly by reading only the opening and closing pages, or label every page to mark document boundaries for Split. An automatic "other" fallback keeps the model from forcing a guess when nothing fits.
{
"output": {
"granularity": "page",
"segments": [
{ "startPage": 1, "endPage": 3, "classId": "invoice", "confidence": 5 },
{ "startPage": 4, "endPage": 5, "classId": "invoice", "confidence": 4 }
]
}
} Decide: Typed answers with probabilities in milliseconds
Runs on decision-machine-1 using CPU encoders with no free-form text generation. Pick from eight typed tasks: yes/no checks, flat or nested classification, rating scales, Q&A, schema extraction, entity tagging and field verification. Every call returns a typed value, a probability score for every option and character offsets for extracted spans at 600 ms p90.
{
"label": "billing",
"probability": 0.995,
"confidence": 0.97,
"scores": { "billing": 0.995, "shipping": 0, "account": 0.005 }
} Gate a branch on a measured probability.
Classify an incoming file, then check whether a claim needs escalation. High-confidence cases move straight ahead; anything below your threshold routes to human review.
Connect your coding agent
Works in Claude Code, Codex and Cursor
- Reads the CloudRaker docs and installs the Paperwork skills.
- Maps where documents, calls and signatures move through your code and process.
- Returns a ranked integration plan and asks before it writes code.
Preview the prompt
Help me integrate CloudRaker, the Paperwork platform, into this project. Read https://docs.cloudraker.com/llms.txt. Append .md to any docs page for Markdown. Install the Paperwork skills for the coding agent you are running in: npx skills add cloudraker/cloudraker-paperwork-skills --agent <client> --skill '*' --yes Choose claude-code, codex, or cursor for <client>. The MCP server at mcp.cloudraker.com serves the whole API. 1. Read this codebase and the process it supports. Find where documents, scans, call recordings, forms or signatures enter or leave it. 2. For each place, name the primitive that fits (parse, split, classify, extract, transcribe, decide, fill, sign) and the surface: the Python SDK, the TypeScript SDK, the paperwork CLI, or MCP. Say which file or step changes and what it replaces. 3. Flag manual steps that one call could replace, even where no code exists yet. 4. Return a ranked list with location, primitive, rationale and effort. Ask me where to start before you write code. 5. API keys are self-serve at https://signup.cloudraker.com, with $15 of free usage a month. Never print keys, and keep them out of the repository and chat. Context: Focus on Decide: Classify files and gate branches on a measured probability, routing low-confidence cases to human review.
Why a decision encoder beats prompting an LLM.
Measured on CPU encoders against general-purpose generative models on routing and verification tasks.
| LLM | Decide | |
|---|---|---|
| Latency | Multi-second generation queues | 600 ms p90 roundtrip |
| Price | Billed on input, reasoning and output tokens | $0.04 per million input tokens, $0 output tokens |
| Uncertainty | Self-reported confidence text | Probability distribution over every label you sent |
| Scope boundary | Can summarize or rewrite | Single pass, stateless, zero free-form generation |
Decide rates and limits.
| Rate | Scope and limits | |
|---|---|---|
| Classify (Documents) | 10 credits ($0.01) / page | 2 to 50 plain-language classes; document mode reads opening and closing pages only |
| Classify (Audio) | 20 credits ($0.02) / minute | Any audio or video file |
| Decide (decision-machine-1) | $0.04 / million input tokens, $0 output | 125M free test tokens/month on every plan; up to 32 texts per batch |
Next in the chain: Act.
Route the output of Classify or Decide straight into Act: fill a form, render a PDF, redact PII or trigger a hosted agent in Run. For full hardware and API specifications, read the decision-machine-1 spec sheet.
Score your first batch of files.
Every plan includes 125 million test tokens a month on decision-machine-1 and $15 of free CloudRaker usage, with no card required.