Ownhand or Idiolect: which should I use?

Pick Idiolect if you want a free beta, a 0 to 100 score for how close a draft is to your voice, or a web studio to draft in; pick Ownhand if you want each draft fitted to one of 12 occasions, replies that match the thread, and a voice that learns from the exact text you send; and if a few lines of tone advice are enough, the free custom instructions in Claude or ChatGPT will do.

By Jordan Gibbs, founder of Ownhand at Thalient Labs. Updated 2 October 2026.

Idiolect facts come from idiolect.app and its GitHub repo as of 2 October 2026. I run Ownhand, so weigh my take with that in mind and follow the links to check it.

The short version

How each one learns your voice at the start

Both start from your own writing. Idiolect asks for 3 to 5 examples, or a short guided interview if you'd rather talk about how you write. Ownhand asks for 5 to 10 things you sent yourself, such as Slack messages, emails and PR descriptions, plus a line or two in your own words. AI drafts don't count as samples, because they teach the model to sound like a model.

Fewer samples means a faster start. More samples give the model more to copy: how long your messages run, how you open them, whether you bother with capitals in chat. Studies back that up. One 2025 study found few-shot prompting gave up to 23.5 times higher style-matching accuracy than zero-shot. Another, covering more than 400 everyday authors, found models still drifted into a generic tone on informal writing even with examples in hand. Samples help a lot, but on their own they leave a gap.

Fitting the draft to where it goes

A Slack DM to a teammate and a first email to a stranger shouldn't read the same, even from the same person. Ownhand makes the setting explicit. Your agent picks one of 12 occasions, from chat_dm to pr_description. Each carries its own register, length, and sign-off, plus checks for that shape. A chat DM is capped at 60 words with no Markdown. A cold email needs a greeting and a sign-off and stays under 125 words.

Idiolect's write and rewrite tools take a format and a situation with each request. Its developer page lists no fixed values for either, so the agent describes the setting in its own words.

That's more open, and it's harder to check whether the draft fits.

Replies

For a reply, your agent passes Ownhand up to 30 recent messages from the thread. Ownhand measures how you write there: message length, lowercase starts, final periods, exclamation marks, emoji, greetings, sign-offs, abbreviations, contractions. The rewrite is checked against those numbers. Facts that only appear in the thread stay out of the rewrite. The thread is never stored.

Idiolect accepts context hints and says they are never stored. Its public docs don't explain how a reply is matched to the thread around it. Try it on a real thread of your own before you rely on it.

Learning from what you send

This is closer than it first looks. Idiolect's refine_writing tool can record what you did with a draft. It has a field for the final text. Its learn_writing_style tool saves writing you agree to share and preferences you state. The docs stop there. They don't say how a final text changes your profile.

Ownhand is built around that step. After you send a draft, your agent calls send_feedback with exactly the text you sent. A reason you give becomes a rule right away. An edit you make without a reason becomes a rule once it shows up twice. In the dashboard you can pin or reject any rule and roll your Hand back to an earlier version. The full loop is in how Ownhand learns from what you send.

Checking a draft

Idiolect's Voice Match scores a draft from 0 to 100 for how close it reads to your own writing. It uses LUAR-MUD, an authorship model from research on telling writers apart. The score is relative. It is not a forensic tool, and its validation holds for the data it was tested on. Use it to compare two drafts, or when a team checks writers against a house voice.

Ownhand gives no score. Every rewrite goes through a checker. It checks for AI-favored words, sentence patterns, the occasion's shape, the thread's style for replies, and whether each link, email address, date, number, and full name in your draft is still there. If a check fails, the model gets the list and tries again, up to two more times. Anything still wrong comes back to your agent with the draft, so you see it before you send.

What a rewrite looks like

These examples are real output from the live API. Each AI draft was rewritten by a demo Hand: a fictional writer (an engineer, a founder, an ops lead, a designer or a consultant) built from a handful of samples.

Two writers worked from the same draft. The engineer dropped the capitals and the follow-up line. The ops lead kept a short opener and turned the ask into a plain statement. Both kept the contract renewal and the 15th. Neither added anything.

Setting up each one

1. Claude Code. Ownhand: claude mcp add --transport http --scope user ownhand https://ownhand.dev/mcp. Idiolect: claude mcp add --transport http idiolect https://idiolect.app/api/mcp. Then run /mcp and sign in.
2. Claude on the web, desktop or mobile. Open Customize, then Connectors, add a custom connector and paste the URL. Idiolect is also in Claude's connector directory. Claude's free plan allows one custom connector.
3. ChatGPT. Turn on Developer mode, then add the URL as an app. OpenAI lists developer mode for Plus, Pro, Business, Enterprise and Education accounts on the web, so a free account cannot add either. Full MCP support is for Business, Enterprise and Edu; Pro connects with read and fetch permissions.
4. Cursor and VS Code. Ownhand has one-click install buttons on Connect, along with setup for Codex, Gemini CLI, Windsurf, Zed, Goose and Cline. Idiolect offers one-click install for Cursor and VS Code.

When Idiolect fits better

• You want to pay nothing for now. The personal plan is free during the beta.
• You want a number for how close a draft is to your voice. Voice Match gives one, and a team can call the Score API.
• You want a web studio to draft in, outside your agent, and a Voice Card you can paste into any chat.
• You have only a few samples, or you would rather describe your writing in an interview.

When Ownhand fits better

• You write in many places, and a Slack DM should read differently from a cold email or a PR description.
• You reply to threads a lot and want the reply to match the people in it.
• You want your voice to keep improving from what you actually send, with rules you can read, pin and reject.
• You want names, numbers and dates checked before a draft reaches you.

Privacy, side by side

Idiolect's privacy policy says it does not use your private writing to train public models. It keeps samples and profiles while your account is active and for a reasonable period after. It deletes them when you delete your account or ask. It uses model providers such as OpenAI to generate and analyze text.

Ownhand doesn't sell your data or train its own models on it. Learning runs on a model through OpenRouter's contributor tier, where the provider may use what it receives to improve its models. That covers your feedback and, when your style card is rebuilt, your samples. Drafts you only rewrite never go there. See is Ownhand safe and the privacy policy.

If you need neither

Both tools fit people who send a lot of messages through an agent. If you write most things yourself and only want the model to sound less stiff, start free. Put a short style note in your custom instructions (see Ownhand vs custom instructions). Learn the phrases that make writing read as AI.

Follow-up questions

Can I use Ownhand and Idiolect at the same time?

Yes. Both are remote MCP servers, so you can connect both to an agent and pick which one to call. Claude's free plan allows one custom connector though. Idiolect is also listed in Claude's connector directory, so check how your plan counts it before you add both.

Is Idiolect free?

The personal plan is free while it's in beta. No card needed. Idiolect says it will publish paid plans on its pricing page first and email existing accounts. For team and API use, including the metered Score API, contact them for pricing.

Does Ownhand give a voice score like Idiolect's Voice Match?

No. Ownhand checks each rewrite against a word list, sentence patterns, the occasion, the thread for replies, and the facts in your draft. When a check fails, it tries again. It tells your agent about anything left. You get a pass or a list of problems, never a number.

Does either one train AI models on my writing?

Idiolect says it does not use your private writing to train public models. Ownhand does not train its own models on your data. Its learning model runs on a provider tier. The provider may use your feedback and samples to improve its models. Drafts you only rewrite never go there.

Sources

Checked on 2 October 2026. If something here is out of date, email hi@thalientlabs.ai and we will fix it.

• Idiolect: developers: remote MCP over Streamable HTTP with OAuth 2.1, Score API, Voice API
• Idiolect: MCP connector: tool list and parameters, setup for Claude, ChatGPT, Cursor, VS Code and Claude Code
• Idiolect: pricing: personal plan free during the beta, Team and API by contact
• Idiolect: Voice Match: 0 to 100 relative score, LUAR-MUD scorer, validation limits
• Idiolect: privacy policy: retention, no training of public models on private writing, model providers
• moshpit-labs/idiolect-mcp on GitHub: profile from 3 to 5 examples or a guided interview, MIT license
• Rivera-Soto et al., Learning Universal Authorship Representations (EMNLP 2021): the LUAR authorship model
• Wang et al., LLMs still struggle to imitate the implicit writing styles of everyday authors (Findings of EMNLP 2025): style imitation from a few examples, across more than 400 authors
• How well do LLMs imitate human writing style? (arXiv, 2025): few-shot prompting gave up to 23.5 times higher style-matching accuracy than zero-shot
• Claude Help Center: custom connectors using remote MCP: Customize > Connectors; free users get one custom connector
• OpenAI: ChatGPT developer mode: Plus, Pro, Business, Enterprise and Education on the web
• OpenAI Help Center: developer mode and MCP apps: full MCP for Business, Enterprise and Edu; Pro with read and fetch permissions
• Ownhand source code: the occasions, the checker, the reply fingerprint and the feedback loop