Inference is replaceable.
Governance is not.
Upload your documents. Ask questions in plain English. Get cited, grounded answers from your own content, with confidence scoring, refusal enforcement, a full audit trail and multi-tenant support.
Answers you can defend.
- RBAC pre-filter
- Maker-checker workflow
- Full audit trail
- PII detection
- Citation enforcement
- Multi-tenant support
- Confidence scoring
- RaggrKairo routing
Governed answers for regulated work
Organisations sit on thousands of policies, circulars and manuals that staff struggle to search. RAGGR is designed to give cited answers from your own documents, with role-based access, an approval workflow and an audit trail.
Answers from general knowledge rather than from your own documents.
Grounded in documents; governance is built separately.
A governance layer for compliance-first organisations, designed into the engine instead of added afterwards.
Six governance layers, by design
Each layer sits inside the engine, not bolted on as an afterthought plugin.
Role-based filtering is designed to apply before retrieval.
No document goes live unreviewed. Four-stage approval: draft, review, approve, publish, with full status tracking.
Every query is logged: user, document pool, model used, answer, confidence score and timestamp. Exportable for review.
Aadhaar, PAN, Passport, IFSC, UPI and Voter ID flagged at ingestion and before delivery, using a PII-detection engine with India-specific recognisers.
Answers are designed to cite their source page.
One engine, many organisations. Separate knowledge pools per organisation, with per-tenant API keys, model overrides and usage dashboards.
Model choice, measured on an internal question set
Internal evaluation set: 130 questions across MCQ, fill-blank, keyword and open-ended formats, run on a technical-course corpus. RaggrKairo routes by query type, not by preference.
| Metric | Cloud model | Open 70B model |
|---|---|---|
| Citation Rate | 82% | 96% |
| Faithfulness | 94% | 96% |
| Avg Latency | 3.8s | 1.2s |
On this internal set the open-weight model had a higher citation rate and faithfulness and lower latency. Internal evaluation on one corpus, not an independent audit; results vary with your documents.
Three ways to deploy
From shared SaaS for a proof of concept to on-premise deployment for data-residency needs.
- Up to 3 document pools
- Shared RAGGR infrastructure
- Full guardrail stack included
- RaggrKairo model routing
- Standard audit logging
- Guided onboarding
- Unlimited document pools
- Dedicated Azure Container Apps
- Custom domain + whitelabel UI
- Model override per tenant
- Enhanced audit + compliance export
- Support by agreement
- Deploy inside your Azure VNet
- Local inference option
- Designed to keep data in your environment
- Full white-label option
- Custom PII recogniser library
- Deployment documentation
Your documents deserve
a governed answer engine
Tell us about your documents and who needs to ask questions of them. We will set up a pilot around that.
Figures come from our internal evaluation, not an independent audit.