Introducing the Evidoc API: Verifiable Document Intelligence for Your Product

The product is now a platform

Until now, Evidoc was a place you went to: upload your documents, ask questions, get cited answers. That’s still the easiest way to use the product, and it isn’t going anywhere.

But over the last few months, we kept getting the same question from teams building their own AI products:

“Can we use your retrieval engine inside our app?”

Today the answer is yes. The Evidoc API is now available — the same Knowledge-Graph reasoning engine that powers our web app, exposed as a server-to-server API behind a managed gateway.

What you get

A single endpoint, POST /hybrid/query, that takes a natural-language question and returns a synthesized answer with:

Citations — every claim is linked back to the specific passage in the source document
Evidence path — the reasoning chain across documents (especially valuable for multi-hop questions)
Route metadata — which retrieval strategy was used and why
Token accounting — exact token usage for cost attribution

Here’s the smallest possible example:

curl -X POST "https://graphrag-apim-wg3temevssbja.azure-api.net/graphrag/hybrid/query" \
  -H "Content-Type: application/json" \
  -H "Ocp-Apim-Subscription-Key: <YOUR_KEY>" \
  -H "X-API-Version: v2" \
  -d '{"query": "What is the contract termination notice period?"}'

That’s it. No SDK to install, no auth dance, no local setup. The same retrieval engine that powers Evidoc.com is now one HTTPS call away.

Three reasoning routes, picked for the question

Different questions need different retrieval strategies. The API exposes three:

hipporag2_community — the default. Best for single-document or topic-coherent questions. Fast and accurate.
hipporag2_experimental — for multi-hop reasoning where evidence is scattered across documents. “Do these invoices match the contract amendment?” — the kind of question that requires chaining facts across files.
hipporag2_coverage — for comparative or aggregative questions. “Compare liability caps across all our vendor contracts.” — wide retrieval, exhaustive synthesis.

If you don’t pick one, the server picks for you based on the query. If you do, you take control of the cost/quality trade-off.

How we secured it

Sending raw API calls to a backend is easy to ship and easy to regret. We chose a different design: every customer call goes through an Azure API Management gateway that we control end-to-end.

The gateway:

Authenticates the caller via a subscription key
Enforces per-subscription throttling
Injects a managed-identity Bearer token before forwarding to the backend over a private channel

You never handle Azure AD tokens. You never expose backend URLs. Your subscription key is the only secret to manage. If we need to rotate keys, change backend topology, or add a new region — that’s our problem, not yours.

For customers who need it, the same gateway supports per-subscription isolation, dedicated throughput, and Virtual Network integration as add-ons.

Who this is for

The API is built for teams who want verifiable answers from their own document corpus, exposed inside their own product. Some of the integrations we’ve seen take shape:

Legal-tech platforms — embedding contract Q&A into their existing review workflow
Customer-support tools — answering tier-1 questions from a curated knowledge base, with citations a human can audit before sending
Compliance dashboards — surfacing policy citations next to AI-generated risk summaries
Internal AI agents — giving an LLM agent a “search my company’s documents” capability that returns cited facts instead of hallucinations

If your product needs an LLM to read documents and never make things up, you’ve found the right tool.

Pricing — prepaid credits, pay as you go

API pricing is prepaid and usage-based. You buy credits up front and spend them as you go — no subscription, no monthly commitment, and credits never expire. Every API call draws down your balance based on real usage, so a simple lookup costs a few credits and a complex multi-hop query costs more.

What that looks like:

Start for $1 — the entry pack is 1,000 credits (≈ 26 queries or 16 pages indexed). Larger packs scale linearly up to $100 for 110,000 credits (+10% bonus).
A typical query costs about 38 credits (billed by measured usage, so simple queries cost less); indexing one page of a document costs 60 credits (exact).
Indexing and querying draw from the same balance — there’s no separate indexing invoice to reconcile.
Dedicated capacity — per-tenant keys with isolated throughput, committed-volume discounts, SLAs, and VNet integration are available for production deployments.

Start for $1 → · Talk to us about enterprise →

What’s next

The full reference is in the Programmatic API Reference PDF — endpoint details, all request/response schemas, error codes, Python and cURL examples. It’s the same doc we hand to integration customers.

On our side, the roadmap for the rest of the quarter:

Self-serve key rotation — rotate keys directly from the dashboard
Per-tenant rate limit isolation — currently shared, soon dedicated
Streaming responses — for chat-style integrations that need first-token latency
Webhook events — for indexing completion and quota alerts

If any of those land on your critical path, tell us — early API customers shape what we build next.

Want to integrate Evidoc into your product? Get your API key — start for $1, or email support@hulkdesign.com with questions.