The right model per page.
Automatically.
Landing AI gives you their model. Docira classifies each page, routes to the cheapest provider that can handle it, and verifies quality before returning. Same accuracy ceiling, 40–70% lower cost, with vendor failover and self-hosted fallback.
One API, nine vision-model providers
What's inside
Built like a router, not a single model.
Four primitives stacked on a multi-vendor backbone. Each one replaces an entire dependency on a single foundation model.
01 — Routing
Multi-model router
We classify each page on complexity (tables, handwriting, math, layout) and route to the cheapest model that can handle it. Fast, Pro, and Expert tiers across 9 providers, with sub-200ms decisions.
02 — Schema
Schema-guided extraction
Pass a JSON Schema; get structured output that conforms to it. Provider-agnostic — the same schema works whether the page lands on Claude, Gemini, or Llama. A/B providers per schema and pick the best.
03 — Quality
Verified accuracy
Every page gets a confidence score against an OCR baseline. Low-confidence pages auto-reroute to a higher tier — no silent quality drops. You see the score on every response.
04 — Reliability
Vendor failover
Per-provider circuit breakers detect outages and reroute around them. One vendor rate-limited or down? Traffic shifts automatically. Your pipeline keeps running on Anthropic, Google, OpenAI, Together, AIMLAPI, Groq, Fireworks, NVIDIA, or your own self-hosted vLLM.
How it works
Three calls. No model lock-in.
POST your document
Send a PDF, image, or office document to /v1/parse — sync (≤25 MB) or async batch.
Router classifies + dispatches
Each page is scored on complexity. Simple pages go to the fastest provider; complex pages escalate to a premium model.
Verified output
Markdown or JSON returned with per-element bounding boxes and a confidence grade. Low confidence triggers an auto-reroute.
Pricing
Pay for the pages you parse.
Effective rate per 1,000 pages drops as you grow. Overages billed monthly — no surprise overage fees, no commit minimums.
Free
100 pages / mo
- All 3 router tiers
- Markdown + JSON output
- Bounding boxes
- Community support
Starter
2,500 pages / mo
- Everything in Free
- 120 req/min burst
- Email support
- $0.010/page over quota
Pro
15,000 pages / mo
- Everything in Starter
- 300 req/min burst
- Webhook delivery
- $0.007/page over quota
Scale
100,000 pages / mo
- Everything in Pro
- 1,000 req/min burst
- Bring your own provider keys
- $0.005/page over quota
Enterprise
750,000 pages / mo + SLA
- BAA + HIPAA-aligned controls
- IP allowlisting + private networking
- Dedicated support, 99.9% SLA
- $0.003/page over quota
FAQ
Things people ask before signing up.
Is Docira HIPAA-compliant?+
Docira is built on HIPAA-eligible infrastructure (Neon Scale, Stripe). The Enterprise plan includes a Business Associate Agreement (BAA), IP allowlisting, and private networking. Free, Starter, Pro, and Scale plans should not be used to process Protected Health Information (PHI).
Are you SOC 2 certified?+
A SOC 2 Type II audit is in progress. We expect attestation later this year. We can share our security questionnaire and current control documentation under NDA — email hello@docira.io.
What document formats are supported?+
PDF, PNG, JPEG, TIFF, DOCX, and XLSX out of the box. Limits per request: 25 MB and 500 pages on Pro+, smaller caps on Free/Starter.
Can I bring my own provider API keys?+
Yes — on Scale and Enterprise plans. We'll route through your OpenAI, Anthropic, Google, Groq, Together, Fireworks, NVIDIA, or self-hosted vLLM keys at pass-through cost. Your keys, your rate limits, your compliance posture.
How does pricing work?+
Each tier includes a monthly page allowance. Overages are billed monthly at the tier's per-page rate (lower per-page on higher tiers). No commit minimums; cancel anytime; included pages don't roll over.
What happens to my documents after parsing?+
Documents and outputs are cached for 30 days for retry/replay, then deleted. On Enterprise we can disable caching, route through your S3 bucket, or sign a custom DPA.