Kosha Copilot (help)
Starter+In-app assistant that explains screens and guides users step by step.
Kosha AI
19 capabilities, one governed AI layer. Kosha AI reads, classifies, summarises and checks so your people can decide faster. It never decides for them, and it never sees what the user cannot.
Ask the library
Ask across every document you are allowed to open. Each answer links to the document, page and passage it came from, so checking takes a click — not an afternoon.
Every AI capability
In-app assistant that explains screens and guides users step by step.
One-paragraph summary and key points for every document.
Suggests class, title, number, revision and metadata from file content on upload.
Ask questions about a document; answers cite page and passage.
Ask questions across every document you can access, with citations.
Find documents by meaning, not just keywords; hybrid ranking.
Explains what changed between revisions in plain language.
Checks the document against the stage checklist and drafts review comments.
Groups duplicate comments and drafts responses for the comment resolution sheet.
Prioritised summary of your tasks, deadlines and what changed.
Extracts text from scanned PDFs and images so they become searchable.
Transcribes recordings with timestamps so meetings and site videos are searchable, and captions photos and drawings with a vision model (within your AI policy).
Translate documents and summaries between languages.
Detects personal and commercially sensitive content and suggests a classification.
Describe a report in plain words — Kosha builds it.
Predicts delays, suggests assignees by workload and flags bottlenecks.
Finds related and duplicate documents automatically.
Use AI steps (extract, classify, summarise, check) inside automation rules.
Run AI on models hosted inside your own network (Ollama, vLLM or any OpenAI-compatible server), so nothing leaves your premises. Included in Sovereign; a licence add-on for other editions.
Governance
Most AI rollouts stall on one question: where does our data go? In Kosha, you decide — per organisation, and per classification level.
Kosha’s built-in on-device engine (nothing leaves the server), a local model on your own GPUs via Ollama or any OpenAI-compatible server, or a cloud provider — Anthropic, OpenAI or Azure OpenAI — under your own agreement. Chat and embedding providers are configured separately.
Pick the highest classification that may be sent to an external provider. Documents above it are only ever processed locally — or not at all.
Optionally require users to confirm, before AI processes a document, that it contains nothing that must not be sent to the configured service. Every attestation is recorded.
Retrieval runs against the same lifecycle-driven access that governs the library, at the moment you ask. AI can never show you a passage from a document you cannot open.
Usage is metered in credits and pooled across your organisation, with a monthly limit you control. No surprise invoices.
Your content is never used to train models — ours or anyone else’s. If you connect a cloud provider, it runs under your own agreement — we recommend zero-data-retention terms.
01
On-device heuristics, OCR and search. No data leaves the Kosha server. Works in the browser demo.
02
Ollama or any OpenAI-compatible server on your GPUs. Full generative AI, fully air-gapped. Included with Sovereign.
03
Anthropic, OpenAI or Azure OpenAI under your agreement, with a classification ceiling and optional attestation.
Predictable cost
Every cloud edition includes a pooled monthly allowance — 100, 500 or 1,500 credits per user on Starter, Business and Enterprise. Sovereign runs AI on your hardware, so there is no credit limit.
Need more? Credits are $6 per 1,000 (₹399), or prepaid packs that roll over for 12 months. Administrators get alerts at 75%, 90% and 100%, and choose between a hard cap, slower fallback models, or automatic top-up. See pricing.
| Action | Credits |
|---|---|
| Page classified, extracted or OCR-indexed on upload | 1 per page |
| AI chat answer (document or library) | 2 |
| Document summary (up to 50 pages) | 3 |
| AI agent run in an automation (shown before it runs) | 10–25 |
| Translation | 1 per 1,000 characters |
| Semantic re-index after an edit | Free |
FAQ
The live demo runs the real Kosha engine inside your browser — no sign-up, and nothing you upload leaves your machine.