
AgentDock
AgentDock markets an AI customer engagement platform due in Q3 2026, but what ships today is Dock: an AI writing editor with verifiable citations, a Chrome extension, an open prompt library and an agent API.
What is AgentDock?
AgentDock presents itself as an AI-native Customer Engagement Platform, but the two halves of that promise sit on different timelines. The engagement product, an AI employee answering customers on web, email, phone, SMS, Telegram and WhatsApp on top of a shared customer memory layer, is announced for Q3 2026 and reachable only through a waitlist. What is delivered today is Dock, described in the documentation as "a writing editor, a Chrome extension, an open prompt library, and an agent API. All live today."
The editor is a document workspace built around tracked changes. Every AI suggestion arrives as a proposed revision that the writer accepts or rejects, and the version history signs each revision as human or agent, naming the model used. Citations come from an academic index rather than being generated: real titles, authors and verifiable DOIs, added by DOI, URL, PMID or arXiv ID, in more than 10,000 styles, with imports from Zotero and Mendeley. Documents export to Word, PDF, LaTeX, Markdown, HTML and plain text, and Word exports keep the revision marks.
The Chrome extension carries the same layer to the rest of the web: a button in every input field, a side panel that reads the current page or a YouTube video, a modal editor over the site, and drafts kept per site. Fourteen built-in actions are grouped into Writing, Editing, Style, Research and Translation, alongside your own prompts.
The prompt library is open, 1,642 templates that an agent can read over MCP, CLI or REST without authenticating. The agent API is the third door: 46 CLI commands (npm i -g agentdockai), 38 MCP tools and 36 REST routes under /api/d/*, covering eight areas of operations including 15 citation operations, and free for agents. An agent loads a SKILL.md before writing, which carries house constraints, style, document templates and the definition of "ready to publish", and it can work in suggest mode, proposing tracked changes for a human to accept or reject.
The publisher also runs its own research, SteerBench-Work and SteerBench-Missions, on when an agent should act and when it should hand back to a person, and keeps its code on GitHub.
What it does
- Draft a full document in a real editor from a single line describing what you want to write
- Attach verified citations by DOI, URL, PMID or arXiv ID, in more than 10,000 styles
- Proofread, rewrite, shorten or expand text in any input field on the web through the Chrome extension
- Summarise the page you are reading, or a YouTube video, from the extension side panel
- Run one of the 1,642 library prompts straight into the editor
- Have an agent write, cite and publish documents through the CLI, the MCP server or REST
- Publish a document to a public /p/{slug} URL with SEO metadata
When to use AgentDock / When not to
A quick filter to help you decide if AgentDock is the right fit.
When to use AgentDock
- Researchers, PhD candidates and students writing a thesis, a literature review or a grant application, who need references attached by DOI, PMID or arXiv ID rather than generated
- Writers and journalists producing sourced articles, who check references and lift pull quotes without leaving the draft
- Developers and agent teams who want an agent to write, cite and publish documents through the CLI, the MCP server or the REST API, from Claude Code, Cursor, Codex, Aider or Cline
- Marketers and community managers drafting newsletters, LinkedIn posts and cold emails from a ready-made prompt library
- Job seekers tailoring a CV to a specific advert and writing the covering letter that goes with it
When not to use AgentDock
- Mobile-first users and anyone working outside Chrome: there is no iOS or Android app, and the browser extension is announced for Chrome only
- Support and sales teams that need the Customer Engagement Platform now: access is by waitlist and delivery is announced for Q3 2026
- Teams looking for shared seats: organisation plans for the Customer Engagement Platform are not available yet
- Buyers whose procurement requires SOC 2, ISO 27001 or a signed data processing agreement, none of which is published
- Non-English speakers: the site and the interface are in English only, with no localised version offered
How to use AgentDock
A typical end-to-end flow, from setup to results.
- Create a free account; the seven-day Unlimited trial needs no payment card and drops to the Free plan afterwards
- From the /write entry point, type one line describing what you want to write, or start from one of the 1,642 templates
- The editor opens with the prompt fields waiting to be filled in
- Write, then trigger an AI action on a selection: improve, rewrite, expand, shorten or proofread
- Accept or reject each suggestion, which arrives as a tracked change
- Add sources by DOI, URL, PMID or arXiv ID, or connect Zotero or Mendeley, then set the citation style for the document
- Export to Word, PDF, LaTeX, Markdown, HTML or plain text, or publish to a public /p/{slug} URL
- Install the Chrome extension to find the button, the side panel and the prompt library in every input field on the web
- For agents: install the CLI with npm i -g agentdockai, get an API key, then point the MCP server at Claude Code or Cursor
- Manage plan, payment method and invoices in the billing settings, handled through Stripe
Pros & Cons
Pros
- Citations are verifiable rather than generated, with DOIs and bibliographic connectors
- No AI edit is applied without human approval, since every suggestion is a tracked change
- Flat pricing with no usage metering and no per-seat billing, at USD 12 per month on annual billing
- Permanent free plan, plus a seven-day Unlimited trial that asks for no payment card
- Three entry points for agents (CLI, MCP, REST), announced as free for agents
- Explicit policy of not training models on user content
- No lock-in: exports to Word, PDF, LaTeX, Markdown, HTML and plain text, and account deletion available
Cons
- The product the site sells hardest, the Customer Engagement Platform, is not available: Q3 2026, waitlist only
- No legal entity, no postal address and no jurisdiction are published, and the terms of use still show a "[jurisdiction]" placeholder
- Generic privacy policy with no effective date and no retention period
- No data processing agreement, no sub-processor list and no certification such as SOC 2 or ISO 27001
- Availability messaging contradicts itself: the /dock FAQ says "launching soon" with a waitlist while the documentation says "All live today"
- No mobile application; the browser extension is Chrome only
- No team or organisation pricing, and the interface and site are in English only
Pricing & Plans
A permanent free plan is available at no cost. The lowest paid entry point is the Unlimited plan at USD 12.00 per month when billed annually (USD 144.00 per year); the same plan is USD 39.00 per month without annual commitment. A seven-day Unlimited trial is offered without a payment card. Agent access through the API, the CLI and the MCP server is announced as free. No organisation or team plan is available at this date.
- USD 0
- free forever
- no payment card required. Unlimited documents
- 10 AI writing actions per day
- the Chrome extension and the full prompt library
- USD 39 per month
- or USD 144 per year billed annually (USD 12 per month). Everything in Free without the daily cap
- unlimited AI writing
- unlimited citations in 10
- 000+ styles
- priority support
- free for agents through the API
- the CLI and the MCP server
- announced
- not available yet
Data, GDPR & hosting
A consolidated view of how AgentDock handles your data.
GDPR overview
GDPR implementation is minimal and appears in one place only, the product documentation rather than the legal pages: "Dock supports the data rights GDPR requires, including access to your data and deletion of it." The privacy policy never names the regulation; it grants a right to "access, update, or delete your personal data at any time" plus an opt-out from marketing communications. The terms of use do not mention it at all. There is no Article 27 EU representative, no named data protection officer, no data processing agreement, no sub-processor list, no stated legal basis for processing, no retention period and no documented international transfer mechanism. The privacy policy carries neither an effective date nor a last-updated date, and the terms of use still contain an unfilled placeholder: "governed by the laws of [jurisdiction]".
Who owns the data?
Under the terms of use you keep ownership of the content you post, but you grant AgentDock a licence to use it, and the published text does not bound that licence in scope, duration or purpose. Content the publisher judges inappropriate may be removed at its discretion, and everything belonging to the platform itself, trademarks and materials included, stays its property. The product documentation is firmer about writing: "Your writing is yours, it is not used to train AI models, and you can take it out or delete it whenever you choose." Documents sit on Dock's servers, tied to your account. No sub-processor list is published, so who else may handle that content is never stated.
Reuse rights
Nothing in the published terms restricts commercial reuse of the text you produce with Dock: the output is yours to publish, sell or repurpose without asking permission. Getting the work back out is just as open, and the documentation is explicit that "you are never locked in": "Any document exports to Markdown, HTML, plain text, or Word, or copies to the clipboard, so your writing can leave Dock the moment you want it to." The product page adds PDF and LaTeX, and Word exports keep the tracked-change marks. A document can also be published straight to a public /p/{slug} URL with SEO metadata. Individual documents can be deleted from the editor, and the whole account from settings.
Data retention & training
Hosting summary
No hosting jurisdiction is published. The documentation says only that "Documents live in your account on Dock's servers, available wherever you sign in and syncing as you write". No country, no region and no cloud provider is named anywhere on the site, and no data processing agreement, sub-processor list or certification such as SOC 2 or ISO 27001 is available. The site itself is served behind Cloudflare (AS13335, anycast address 188.114.96.0), which says nothing about where documents are stored. Declared security measures stay generic: "We implement various security measures, including encryption and secure storage, to protect your personal information." Zotero and Mendeley logins are described as protected credentials. In the Chrome extension, per-site drafts and preferences stay in the browser local storage, while prompts and documents live on the account and are described as encrypted. The privacy policy mentions sharing with trusted third-party service providers; Stripe and Google Analytics are named elsewhere on the site, but no named list of processors is published.
Things to keep in mind
Risks and trade-offs to weigh before adopting AgentDock.
- The publisher is not identified: no legal entity, no postal address and no country of jurisdiction are published, so there is no way to anticipate what recourse a dispute would leave you
- The terms of use are unfinished, since "These terms are governed by the laws of [jurisdiction]" leaves a placeholder inside a contractual document
- The privacy policy is generic, undated and silent on retention, so there is no way to know how long your documents are kept
- The licence granted over posted content is broad and undelimited in the terms of use
- Automatic citations still have to be checked by the author: a valid DOI proves that a reference exists, not that it supports the sentence it is attached to
- Handing writing over to agents erodes, run after run, both command of the text and editorial responsibility for it
- The flagship product is a Q3 2026 announcement, so committing to it means trusting a roadmap rather than a shipped product
Setup & Integrations
Technical difficulty
Two very different levels. For everyday writing there is nothing to set up: create an account and start, with no installation and no configuration. The Chrome extension adds a single install, with its permissions listed up front (scripting, side panel, tabs, cookies, storage, context menus, identity). Agent use is a developer task: install the CLI through npm, obtain an API key, then configure the MCP server for Claude Code, Cursor, Codex, Aider or Cline, each of which has its own documentation page. Non-technical users can be writing in minutes; connecting an agent assumes a working command line.
Deployment
Integrations
Supported languages
Behind AgentDock
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement AgentDock.
Frequently asked questions
What can I actually use today?
Is there a free plan?
How much does the paid plan cost?
Is there a free trial?
Is my writing used to train AI models?
Where do the citations come from?
Can I get my documents out?
Is there an API?
Is there a mobile app?
Is there a team plan?
Should you pick AgentDock?
AgentDock sells one product and delivers another. The headline promise is an AI-native Customer Engagement Platform, an AI employee answering customers across web, email, phone, SMS, Telegram and WhatsApp, and that product is announced for Q3 2026, behind a waitlist. What exists today is Dock: a writing editor, a Chrome extension, an open library of 1,642 prompts and an agent API, all on one account. Read the offer with that gap in mind and the tool becomes much easier to judge.
On its own terms, Dock is a serious piece of work. Citations are drawn from an academic index and attached by DOI, URL, PMID or arXiv ID instead of being invented, with Zotero and Mendeley imports and more than 10,000 styles. Every AI suggestion arrives as a tracked change that a human accepts or rejects. Pricing is flat: a permanent free tier, then USD 12 per month on annual billing, with no per-seat or usage metering, and agent access through the CLI, the MCP server and REST is free. The publisher states that user content is not used to train models, and exports to Word, PDF, LaTeX, Markdown, HTML and plain text mean nothing is trapped.
The weak point is not the product but the company behind it. No legal entity is named, no postal address is published, and the terms of use still read "governed by the laws of [jurisdiction]", an unfilled placeholder inside a contract. The privacy policy is generic, undated and silent on retention. The domain was registered on 01/12/2024 and first archived on 07/01/2025, and two co-founders are named. Useful for sourced academic and professional writing, and for teams that have agents write, provided that legal opacity is acceptable.
- Choosing a selection results in a full page refresh.
- Opens in a new window.