How Dokeo works
AI writes. Rules decide. You approve.
Dokeo is the control layer between AI-made work and the moment it ships. Models write the fix, rules decide pass or fail, and a person approves what goes out. Below are the five layers, and each line is marked live, on request, or roadmap.
- Live works today, for any workspace
- On request built, given to a workspace on request
- Roadmap not built, or not connected to customer data yet
Rules decide
Deterministic engine
The part that says pass or fail. No model is in the path, so the same text always gets the same verdict.
19 core checks across SEO, AEO and GEO
LivePlus type-specific rules for 17 content types, so a changelog is not judged like a landing page.
DocsSame text, same verdict, provable later
LiveEvery scan returns a decision hash over the text, your settings, the engine and your rules. Replay it and Dokeo says whether the decision is identical, or which input changed.
DocsRules versioned per workspace
LiveBrand rules and rule packs carry a version, and the version is part of the decision hash.
DocsBenchmark snapshots
RoadmapMarket benchmarks stored as dated, hashed snapshots, so a score that moves can be explained and replayed.
AI writes
Intelligence engine
Models draft the fix. They never decide the verdict, and the gate re-scores whatever they write.
Suggested fixes, one check at a time
LiveShown as a diff. Nothing is saved until you accept, and the same checks run again on the result.
DocsFix until it passes
LiveRewrite, rescan and repeat up to three times. A passing draft goes to review, never straight to publish.
DocsSensitive checks stay out of models
LivePersonal data, confidential text, claims, sources and duplicate content are never sent to a model. A person edits those.
DetailsModel-independent
LiveMiniMax by default, with Anthropic, GLM, OpenRouter or your own OpenAI-compatible server as alternatives. The gate does not depend on any of them.
DetailsModel comparison by outcome
RoadmapWhich model fixes best at what cost, scored by the gate itself.
Your rules
Knowledge engine
What makes the gate yours: brand rules, approved examples and the documents your team already wrote.
Brand rules and rule packs
LiveBanned terms, required sections, reading level, citation policy and more, applied to every scan in your workspace.
In the appLearn rules from a style guide or approved examples
LiveDokeo proposes draft rules. A person approves them before they apply.
In the appKnowledge base
LiveUpload brand voice guides, playbooks and benchmarks. They apply to your scans and never to another workspace's.
In the appLive market benchmarks
RoadmapBenchmarks that update on their own. Today they are written into the checks.
One engine
One engine, several surfaces
The same pattern of rules, verdict and approval applies beyond content. Only content is live.
Content
LiveBlog posts, emails, landing pages, docs and more, from the web app, REST API, CLI and webhooks.
DocsAgents and assistants
Live15 MCP tools put the same gate inside Claude and Cursor.
DocsOutbound and GTM
On requestPre-send campaign QA runs through the API and in the app (Outbound check), in shadow mode. It checks each email, keeps a do-not-contact list that never forgets a no, checks send timing, and flags a change of voice between steps. It never sends mail. Available to design partners on request. Sequencer connectors are not built: call the API from n8n, Clay or your sequencer.
DocsCRM and revenue operations
Roadmap44 checks are defined for CRM, sales, RevOps and pipeline. No customer data is connected yet.
Details
You approve
Decision, approval, action, outcome
What happens after the verdict. A person stays in charge of anything that ships.
One verdict: ship or fix
LiveFive severities, one answer per draft, the same in the app, the API and the CLI.
DocsReview queue and approvals
LiveA person signs off before a draft ships. Every decision has an owner.
DetailsSigned webhooks
LiveSend verdicts to Slack, your CMS or a CRM, signed and retried.
DocsGitHub Action
LiveFail a pull request on a bad draft. Public on GitHub as rasulshaikh/dokeo-action.
DocsRelease approved work into your tools
RoadmapPublish approved content into a CMS, CRM or sender from Dokeo. Today Dokeo never sends anything itself.
Outcomes tied back to decisions
RoadmapCitation watch exists after publish. Linking each result back to the decision that allowed it is not built.
What Dokeo will not do
The model drafts, and you decide what ships.
A model never decides pass or fail
The gate has no model in its path. Models only write suggestions, and the gate re-scores them.
DocsYour content does not train a model
Dokeo does not train models on your content. AI providers can keep requests for a limited time, so this is not a zero-retention promise.
DetailsDokeo never sends anything for you
No emails, no posts, no CRM writes. You or your own tools send.
DetailsA person approves what ships
Fixes are suggestions until you accept them, and a passing draft goes to review.
Details