RunInfra is now public.See what's new
RunInfra
Own your AI. We benchmark GPUs, optimize kernels, and deploy open-source models as production APIs.
Start building
Document AI with open vision models.
Open vision-language models parsing PDFs, forms, and tables to JSON. Per-page billed becomes per-million-tokens.
Invoices
Contracts
Forms
Tables
What you actually own
The optimization knobs, the codebase, the model choice. None of it locked away.
01
Per-million-tokens, not per page.
Closed document AI charges per page. RunInfra bills compute on your GPU. Long documents become cheap.
02
Your documents never leave.
Contracts, invoices, and forms stay on your runtime. No third-party retention windows.
03
Tune the schema on your real forms.
Bring your schemas and ground-truth labels. The model fine-tunes on your specific layouts.
Three ways to ship document AI
Most teams pick between speed and control. RunInfra keeps both in one workflow.
| What matters | RunInfraRecommendedFast path with model control and export. | Closed extraction APIsPer-page, hosted. | DIY self-hostingFull control, heavy operations. |
|---|---|---|---|
| 01Launch | Pick model, optimize, deploy Start quickly and keep the production path open. | Call provider endpoint Fast first demo, but the runtime stays rented. | Build serving stack first Infrastructure work comes before product learning. |
| 02Model control | Bring the model ID Keep model choice and serving decisions visible. | Provider catalog You use what the provider exposes. | Your model Full control if your team maintains the runtime. |
| 03Tuning | Measured latency and GPU cost Compare serving choices before deployment. | Opaque Latency and batching stay behind the API. | Manual profiling Your team owns tuning and regressions. |
| 04Export | Managed now, export when needed Use the endpoint first and take the deploy package later. | Locked endpoint You keep calling the provider. | Already owned Export exists because you built everything yourself. |
| 05Operations | Low until you choose to own it Operate managed, then export with the same measured plan. | Low, with lock-in Less infra work, less production control. | High You own infra, failures, upgrades, and serving changes. |
| 06Security | SOC 2 Type 2 Audited controls across access, logging, and incident response. | Varies by vendor Compliance depends on the third party sitting in the request path. | You build it Your team owns the audit trail, logging, and access controls. |
RunInfra
RecommendedFast path with model control and export.
Launch
Pick model, optimize, deploy
Start quickly and keep the production path open.
Model control
Bring the model ID
Keep model choice and serving decisions visible.
Tuning
Measured latency and GPU cost
Compare serving choices before deployment.
Export
Managed now, export when needed
Use the endpoint first and take the deploy package later.
Operations
Low until you choose to own it
Operate managed, then export with the same measured plan.
Security
SOC 2 Type 2
Audited controls across access, logging, and incident response.
Code you own. Deploy anywhere.
The full recipe ships with you. Codebase, kernels, engine config, weights. Run it anywhere.
Live
Add a new field to the extraction schema...
Schemas3
runinfra-document-ai/
samples/
evals/
Managed RunInfra
Our GPUs, per-million-tokens billing from L4 to B200.
Your infrastructure
AWS, GCP, RunPod, bare metal. Same Dockerfile, your cluster.
Local workstation
docker compose up. Full pipeline on a single GPU.
Any HF vision-language model works
Every vision-language model on Hugging Face runs through the same recipe. Search the live catalog above. The examples below are just a starting view.
HF
What RunInfra tunes
Every stage of the pipeline, retuned per model and GPU.
Vision encoder fusion
Tiled patch embeddings fused with the LLM step. Page-shape preserved.
Schema-driven extraction
Output constrained to your JSON schema. Field-level confidence scores.
VLM quantization
AWQ INT4 on the LLM head, FP16 preserved on the vision encoder.
Page-aware batching
Variable-page batches with PagedAttention. No padding waste on long contracts.
Multi-page reasoning
KV cache shared across pages so the model sees the full document.
Pipeline scheduling
Render, extract, validate interleaved per page. Backpressure on overload.
Try this pipeline
Edit the model, engine, or GPU inline. Send to retune the stack in the dashboard.
Common questions
Can't find what you're looking for? Get in touch
What document types are supported?
PDFs, scans, photos, and screenshots. Multi-page documents are processed page by page with consistent output.
