Features

Understand.
Act. Connect.

Most document AI tools do one of these well. Dokko does all three — designed to work together from the first document you index. The same engine serves your internal teams and your customer-facing assistants.

Know

Documents understood end-to-end.

Generic tools chop documents into fragments and index those fragments in isolation. A clause buried deep in a 300-page contract gets indexed without ever seeing the chapters that define its terms. Ask a question that depends on the whole, and you get an answer that sounds right but isn't.

Dokko reads every document in full before indexing any of it. Text, tables, charts, diagrams — multimodal from the start. Every chunk is enriched with the context of the document it came from, so retrieval already understands what a passage means, not just what it says.

.pdf
.docx
.xlsx
.csv
.pptx
.txt
.md
.html

/01 Reads in full, multimodal

The whole document is the unit of truth.

At index time, Dokko reads the entire document — including tables, charts, financial statements, technical diagrams, scanned pages — using a vision-capable model. Every searchable unit is indexed with a summary of where it fits in the document.

Compliance frameworks, legal contracts, technical manuals, regulatory filings — the documents where partial answers are dangerous. Dokko reads the structure, not just the words.

/02

Cross-document, in context.

One question, scoped across an entire knowledge base. Dokko returns the relevant passages from every relevant document, cross-referenced, ranked by their fit to the question.

The real question is rarely "what does this contract say." It's "what do all our supplier contracts say about indemnity in Germany." Dokko answers that as one question — not ten — drawing from every document that actually applies.

One question. Every relevant document, cross-referenced.

/03

Document-level citations.

Every answer cites the source documents it drew from. Click the citation, open the document, read the passage in its original context.

AI you can defend. Verify, audit, and explain every answer to a regulator, a customer, or an internal review board. Citations identify the source document — not a fragment ripped out of context.

Every answer cites the source documents.

/04 Worked example

A European insurer indexes 4,200 pages of underwriting.

The insurer indexes its 4,200-page underwriting manual, 180 binders of historical case decisions, and the live regulatory framework. An underwriter asks: "Has this kind of risk been declined before, and what was the precedent for the decision?"

Dokko reads the question, retrieves passages from the manual, the historical decisions, and the regulation simultaneously, and returns one answer that cites three source documents. The underwriter opens each citation, reads the original, and decides. The whole loop takes 90 seconds — not the 40 minutes the team used to spend searching three separate systems.

Act

Two ways to take action.

Understanding without action is a dead end. Dokko gives you two distinct ways to act on what it finds — and they are not the same thing.

Skills work inside the conversation. Agents work across your stack. You use both, configured by the team that needs them.

Early access

Skills and Agents are in early access — production-usable today, actively iterating with the teams onboarding now. Talk to us about onboarding.

Skills Early access

In-conversation actions, tied to your knowledge base.

A Skill is a configurable rule that activates during a conversation when a user asks about a specific kind of thing. Each Skill is a short instruction the AI follows — and, optionally, a custom Lambda tool the AI can call to fetch live data or run a computation.

Skills are configured per team. The legal team's Skills are not the customer-support team's Skills. They work alongside retrieval, not instead of it — shaping the conversation while the knowledge base does the reading.

Contact directory Skill

The scenario

A team uploads a contacts directory and wants it to be the authoritative answer to "who do I call about X?" — not buried under footer mentions from 50 other documents.

What the Skill does

The Skill activates on contact-style questions, looks up the right record, and returns a structured contact card in the conversation.

Coverage calculator Skill

The scenario

A paint manufacturer's customer asks "how much paint for 50 m²?" Pure document retrieval finds the spec sheet that says "200 g/m²" but can't reliably do the multiplication.

What the Skill does

The Skill activates, pulls the rate from the retrieved document, runs the calculation, and returns "10 kg, based on the data sheet for product X."

Competitor-guardrail Skill

The scenario

A betting operator wants its customer-facing assistant to refuse to discuss competitor brands. Regex blocklists miss variants and indirect references.

What the Skill does

The Skill recognises competitor mentions in context and politely redirects — without breaking on the brand name showing up in a legitimate document.

Skills are activated by a small classifier model that runs in parallel with retrieval — no extra latency. Configure them in the dashboard, no engineering tickets.

Configure Skills in the dashboard

Agents Early access

Multi-step workflows in the tools your team already uses.

An Agent runs a sequence: read this incoming document, extract these fields, compare against this template, draft a response, post it in Slack, file the ticket in Jira, notify the reviewer. Compose once, run continuously. Multi-step. Across the stack.

Agents connect through MCP (native protocol), OpenAPI specifications for the long tail of business systems, and a growing library of hundreds of business tools that already speak those protocols.

Contract intake Agent

The scenario

New supplier contracts arrive at uneven cadence; legal can't review every one immediately.

What the Agent does

Reads new contracts as they arrive, extracts liability clauses, compares against your template, flags deltas in a Slack thread, opens a Jira ticket for review. The legal team sees only the contracts that need attention.

Compliance review Agent

The scenario

Regulators publish updates nightly. Cross-referencing them against your current positions is real work.

What the Agent does

Scans a folder of new filings nightly, cross-references against your current compliance positions, drafts an internal brief for any material change, posts it to a Notion page the team reviews each morning.

Customer-onboarding Agent

The scenario

Every signed agreement triggers the same dozen account-setup steps. People do them wrong.

What the Agent does

Reads the signed agreement, extracts the rate plan, configures the CRM account, drafts the welcome email referencing specific terms, and sends it after a human approves.

Agents are configured visually in the dashboard. Each step is inspectable. Each tool call is logged.

Build an Agent

Connect

Plugs into the tools your team already uses.

Knowledge without action is half a product. Skills and Agents need to reach the tools your team works in every day. Dokko gives them access to a library of business tools, plus standard ways to plug in your own logic.

/05

A library of business tools

What it is

A growing library of hundreds of business tools — Slack, Jira, Zendesk, Notion, Workday, BambooHR, GitHub, Confluence, HubSpot, Salesforce, and the long tail of CRMs, project trackers, document stores, and ticketing systems. Authenticate once via OAuth from the dashboard; every Skill and Agent on your team can call it.

Why it matters

The first integration question is usually "can Dokko talk to X?" For the tools most teams use day to day, the answer is yes — without writing integration code.

/06

Custom Agents via MCP and OpenAPI

What it is

For the systems specific to your business — your internal CRM, your billing platform, your data warehouse — expose an MCP server you control, or point Dokko at an OpenAPI specification. Either way, your Agent gets typed access to the endpoints.

Why it matters

Your internal stack stays in your control. Dokko connects as a client over standard protocols (MCP, OpenAPI), so anything that speaks them slots in.

/07

Custom Skills via Lambda

What it is

When a Skill needs to do something beyond retrieval — fetch a live value, run a calculation, look up a record — attach a Lambda function as the Skill's tool. The Skill calls it inside the conversation; the result flows into the answer.

Why it matters

In-conversation custom logic without rebuilding the assistant. Each team ships the tools its Skills need, in code, on its own cadence.

Two surfaces

Internal and external. Same engine.

Dokko was built for two deployment shapes. Pick one or run both — the engine doesn't care, and you don't pay twice.

For your team

Internal assistant.

Dokko runs as an internal assistant for the legal team, the compliance team, the operations team, the technical team. Each team owns its own knowledge bases, Skills, and Agents.

What teams build

  • Internal contract review assistants
  • Compliance Q&A grounded in current regulation
  • Technical documentation lookup for engineering and support
  • Operations runbooks queryable in the moment of work

For your customers

Customer-facing.

Dokko runs as a customer-facing assistant embedded in your product, your help centre, or your sales site. One script tag. Customer-facing Skills. Live session takeover when an operator needs to step in.

What teams build

  • Customer-support assistants for product Q&A and troubleshooting
  • Sales assistants that read your catalogue and qualify intent
  • Self-service tools for policy, claims, and account questions
  • Onboarding assistants for new users

Two surfaces. One engine. One subscription. The legal team's internal assistant and the company's customer-facing assistant share the same document understanding, the same security posture, and the same dashboard.

/08 Embed anywhere

One script tag. Any website.

Add a script tag with your tenant and config IDs, and your customer-facing assistant is live — no backend changes, no infrastructure to manage. Theme and behaviour customisable; framework-agnostic by design.

<!-- Add to your page HTML -->
<script
  src="https://widget.dokko.ai/version.js"
  data-dokko-tenant-id="YOUR_TENANT_ID"
  data-dokko-config-id="YOUR_CONFIG_ID"
  async
></script>
 
// Initialize the widget
if (window.Dokko) window.Dokko.init();
 
// Optional: control theme
// window.Dokko?.setTheme('dark'); // force dark
// window.Dokko?.setTheme('light'); // force light
// window.Dokko?.setTheme('auto'); // follow system (default)
 
// To remove the widget:
if (window.Dokko) window.Dokko.dispose();

Works with any framework — dark, light, and auto theming built in.

More in the box

Everything else included on every tier.

/01

Multi-model flexibility

Every major AI provider, assigned per-agent from the dashboard. Use the right model for each workflow.

/02

BYOK

Bring your own provider keys. Subscription price stays the same; your provider tokens never leave your account.

/03

Multilingual

100+ languages out of the box. Ask in English, get answers from your German technical manuals.

/04

Web crawling

Point Dokko at a URL set; it strips navigation and ads and indexes only meaningful content. Unchanged pages aren't double-billed.

/05

Embeddable

One script tag, any website. Embed the assistant, the Agent, or the configuration UI.

/06

Audit logs

Every query, every answer, every source, time-stamped. Available from Pro.

/07

Caps you control

Every meter has a published overage rate and a per-tenant policy. Set an optional monthly overage ceiling — your seatbelt against runaway loops.

Get started

Three capabilities.
One trial.

The 14-day trial runs on your real library. Bring your contracts, manuals, or frameworks — and see how Dokko reads them.