Frequently asked questions.

ChatGPT is a general-purpose chat tool that doesn't know your documents, can't be governed, and can't be embedded on your customer-facing surface with the right guardrails. Dokko reads your document library in full, cites the source documents on every answer, deploys both internally and externally, and is configurable per team. ChatGPT is for individuals; Dokko is for organisations that work with documents.

NotebookLM is a personal document tool — one user, one notebook, one session. Dokko is a team and organisation product: multiple users, role-based access, per-team configuration, audit logs, customer-facing embed, and Agents that take action across your stack. If you've outgrown NotebookLM because you need governance, team-wide deployment, or actions on what the AI finds, Dokko is the next step.

Probably yes. Dokko handles PDFs (text and scanned), Word, Excel, PowerPoint, Markdown, plain text, CSV, JSON, HTML, and web pages crawled by URL. It's specifically designed for complex, dense, cross-referenced documents — contracts, compliance frameworks, technical manuals, regulatory filings, clinical guidelines — but it handles simpler content too. The free trial runs on your real documents, which is the fastest way to find out.

Yes — that's what Skills are for. Skills are configurable rules that activate during a conversation when a user asks about a specific kind of thing. Examples: refuse competitor questions, return a structured contact card when asked for the right contact, run a coverage calculation on a value pulled from a spec sheet. Each team configures its own Skills. No engineering tickets.

Yes — Dokko deploys customer-facing as well as internally. One script tag embeds the assistant on your website, your help centre, or inside your product. Customer-facing surfaces use Skills configured for that audience (refusal rules, structured lookups, calculators) and can hand off to a human operator when the AI is uncertain.

Document retrieval — the Know pillar — is generally available and in production at customer sites today, across veterinary, sports, manufacturing, and pharmaceutical regulatory deployments. Skills (in-conversation actions) and Agents (multi-step workflows) are in early access: production-usable today, with teams actively running them, and we're iterating quickly with the customers onboarding now. Pricing, metering, and overage controls described on the pricing page apply now. Surface details may shift while Skills and Agents mature. Talk to product engineering if you want a deeper read on what's stable today and what's actively moving.

A monthly subscription plus a published overage rate per meter. The subscription covers three direct meters — pages, AI tokens, tool actions — each capped per tier. Beyond the cap, you choose per meter whether to pause or allow overage at the published rate. Set an optional monthly overage ceiling and overage-enabled meters auto-pause once it's reached — your seatbelt against runaway agents. Plans start at $99/month with a 14-day free trial. No seat caps. No surprise invoices.

Pages, AI tokens, and tool actions — three direct meters, each with a per-tier allowance. One knowledge base and one data source per tenant on every plan. Skills, Agents, the customer-facing embed, BYOK, and access to every supported AI model are included on every tier. What scales by tier is the allowance volume and the parser quality (Standard vs Premium).

When you bring documents into Dokko, we read them once, understand structure and content, and index them for retrieval. You're charged per page at that point — not on every query. Each plan includes a monthly page allowance (7,000 on Starter, 20,000 on Pro, Custom on Enterprise); overage is $0.03/page at every tier. Before every import, Dokko previews the page count and the cost. You confirm or cancel. Premium parser pages count 5× against the allowance to reflect their higher processing cost.

Each plan includes a monthly token allowance (10M on Starter, 20M on Pro, Custom on Enterprise). Tokens are measured directly — every model SDK reports them — and each AI model has a published multiplier relative to the Haiku 4.5 baseline (1×). Pick the right model for each job and see exactly how your budget depletes. No credits, no opaque AI usage percentage.

For PDFs, one page = one page of the document. For everything else (DOCX, spreadsheets, presentations, Markdown, plain text, CSV, web pages), one page = 2,000 characters of content — roughly a standard page of text.

Yes. 14 days at full Starter capability (7,000 pages, 10M tokens, 500 tool actions, customer-facing embed included, BYOK supported). Credit card required at registration. No charge during the trial; cancel anytime.

Skills are configurable rules that activate during a conversation when a user asks about a specific kind of thing. Each Skill is an instruction the AI follows — and optionally a small custom Lambda tool the AI can call. Skills run alongside document retrieval, not instead of it: the knowledge base does the reading, the Skill shapes the conversation. Examples: refuse competitor questions, return a contact card, run a coverage calculation, flag a low-confidence answer for human review.

Agents are multi-step workflows that run across your stack. An Agent reads, extracts, compares, drafts, posts, and notifies — in the tools your team already uses. Agents connect via MCP (native protocol), OpenAPI specifications, and a library of hundreds of business tools. Examples: read incoming contracts and post redline drafts to Slack; review nightly regulatory filings and brief the team in Notion; configure a new customer's CRM account from their signed agreement.

A Skill changes how Dokko answers inside a conversation — shaping the response, adding a structured lookup, refusing something, running a quick computation. An Agent runs a workflow that goes beyond the conversation — reading a document, taking multiple steps, calling external tools, often with a human approval gate. Most teams use both: Skills shape the customer-facing assistant; Agents handle the back-office workflows.

Dokko is hosted in the EU by default — every standard tenant runs on EU infrastructure. For Enterprise deals, we can provision a separate installation in a different region on request, subject to contractual scoping.

Every answer cites the source documents it drew from. Audit logs (from Pro) record every query, every answer, every source — time-stamped — for compliance and review. Configurable confidence thresholds let you set when Dokko should flag or escalate an answer rather than guess. This is the operational basis of an AI governance posture that survives audit.

SSO isn't shipped yet. When it ships, it will be available on Enterprise. We don't list features on the pricing page before they're real.

Yes — BAA is available on the Enterprise tier. Talk to sales.

A library of hundreds of business tools — Slack, Jira, Zendesk, Notion, Workday, BambooHR, GitHub, Confluence, HubSpot, Salesforce, and the long tail — connectable through one OAuth flow from the dashboard. For systems specific to your business, expose an MCP server or point Dokko at an OpenAPI specification. For custom in-conversation tools, attach a Lambda function to a Skill. Connect once; every Skill and Agent on your team can use it.

Yes — MCP is a first-class protocol in Dokko's agent layer. Agents call MCP servers directly; tenants can author and host their own MCP servers. As the standard matures, new MCP-compatible tools slot in without changes to your Agent configuration.