Frequently asked questions

SHERLOX assists you every day: drafting and replying to your emails, summarizing or comparing documents, analyzing data, preparing your meetings or querying your connected business tools. You describe your need in natural language, and the assistant handles the rest.

Yes. You can attach documents (text, PDF, Word, spreadsheets, images…) and even a .zip archive: the assistant reads the files it contains as if you had attached them one by one, and tells you which ones it could not read. Very long content works better as an attached file than pasted into the message.

Yes. SHERLOX produces Word documents, Excel spreadsheets, PowerPoint presentations and PDFs, and creates images from your description. It can also modify a file you send it. The result can be downloaded directly from the conversation.

Yes, your conversations are kept in your history and available at any time. You can reopen them to continue an exchange or find a previous answer.

Yes, like any AI, SHERLOX can be wrong or lack context. Check important information, especially before making a decision, and don't hesitate to ask it to clarify or correct.

Yes. You can rewind the conversation to the message of your choice and start again from there, without starting over. If it has simply become too long, use compaction instead.

Yes, but it is generous: you can paste a long report or a transcript of several tens of thousands of characters. As you approach the limit, the input box warns you; beyond it, sending is blocked and your text stays in place. For very large content (export, log), attach it as a file instead.

Yes. Ask it to chart figures (bars, lines, areas, shares of a whole) or to place addresses on a map: the visual is displayed directly in the conversation. An address that cannot be found stays listed without breaking the rest of the map.

A traditional chatbot only answers. SHERLOX acts: it connects to your tools, remembers your preferences, runs scheduled tasks (routines) and draws on dedicated skills. It's an assistant that does things, not just one that chats.

In Internal mode, your exchanges rely exclusively on sovereign models hosted in French datacenters: your data never leaves that environment. External mode gives access to more powerful models, suited to tasks that don't involve sensitive data.

Yes. Each agent is specialized (a domain, a set of tools, a role). You choose the agent best suited to your need, and you can use several depending on the tasks.

Yes, if you allow it: nothing is remembered until you turn memory on. SHERLOX then remembers what is useful to you (how you work, a recurring context). You can withdraw your consent at any time and choose to keep or erase what was remembered.

In Internal mode, SHERLOX uses only sovereign models hosted in French datacenters. In External mode, it relies on the market's most capable models for non-sensitive tasks.

Create an agent and give it a name, a colour, an image and a context: the standing instructions it applies in every exchange. Then file the related conversations under it; its page shows its skills, routines and latest activity at a glance.

Yes. Attach skills to the agent along with their documents: it uses them whenever the question relates to them. An agent's library has a limit, which is shown when it is reached.

Yes, with your whole company or through an invitation link you can revoke. Members work in the agent with its instructions and skills, without being able to reconfigure or delete it.

With your permission, autopilot regularly reviews your workspace: it suggests creating agents, filing related conversations under them and drawing reusable skills from them. You accept, adjust or reject each suggestion; if it finds nothing useful or you ignore its suggestions, it steps back.

Just tell it, for example "remember that I sign my emails Best regards". The fact is saved immediately and pinned: it will never be reworded or merged with another, and saying it again does not create a duplicate. Memory must be turned on.

A skill is a ready-to-use know-how that you add to your agent: a procedure to follow, a response format or an expertise that the assistant applies automatically when the need arises.

Yes. You choose a skill's scope: just for you, for the whole company, or for the people you name. A skill the assistant writes during a conversation stays private until you ask to share it.

The assistant applies the skill that matches your request on its own. You can also call it explicitly: type "/" in the input box and pick it from the list.

Yes. When you explain a procedure to it during a conversation, it can offer to save it as a skill of the agent concerned. The skill stays private until you ask to share it.

A routine is an automated task that runs on a schedule (every morning, every week, etc.). For example, receiving a summary of your unread emails or of your day at a set time.

Describe the task to automate and its frequency to the assistant, or set it up from the dedicated routines area. You can then edit, pause or delete it at any time.

You are notified in the app and on your device, with the reason for the failure (disconnected tool, unavailability…) and a link to the routine's conversation. Fix the cause, for example by reconnecting the tool concerned: the routine will run again at its next scheduled time.

Yes. Ask in one sentence ("every Monday at 8 am, give me an overview of my meetings for the week"), or let it suggest a routine when a request keeps coming back. A routine written by the assistant says so, and their number per agent is capped.

All runs of a routine are grouped in a single conversation, each with its own messages: you follow its history without one conversation per day. An agent's routines also appear on its page.

For each routine you can pin the account it should use, regardless of the one you use day to day. If you do not choose, it uses your usual account.

To fine-tune it. If you answer yes, it can shorten the routine's wording for the next runs; if you answer no, it takes you back to the routine so you can correct it.

Yes. Attach your Word, Excel, PowerPoint or PDF file and describe the change. You get your document back, modified where you asked and identical everywhere else. You can also point to a specific area in the preview.

Yes. Attach the PDF or Word template (contract, purchase order, form) and give the information to enter: the assistant finds the fields to fill in and returns the completed document, identical everywhere else.

Yes. Provide your PowerPoint or Word template: the document produced reuses its masters, layouts, branding and dimensions.

Yes. The workbooks produced contain real dates, real numbers and their formats, not text that looks like them: you can sort, filter and calculate straight away.

Yes. Once the file has been produced, ask the assistant to send it to your recipients: it goes out as an attachment from your connected mailbox, whatever its type.

The draft is shown as a card: review it, then send it, save it to your drafts or copy it. Ask for another version if the tone or content does not suit you.

At the end of a meeting, the assistant takes the transcript, draws up the minutes and emails them, to the organiser only or to all participants depending on your company's setting. The transcript remains accessible from the email, then disappears with the minutes at the end of the retention period.

From your settings, choose the tool to connect (mail, calendar, storage, business tools, etc.) and authorize access. Once connected, the agent can act on that tool on your behalf.

It means the tool needed for your request is not yet linked to your account, or the authorization has expired. Reconnect the tool from your settings so the agent can use it.

Yes, at any time from your settings. Once disconnected, the agent no longer has access to that tool and the routines that depend on it are suspended.

Your account is connected, but it does not give you the right to perform this action in the tool (viewing another team's data, for example). Reconnecting will not change anything: ask the tool's administrator for the missing permission.

Yes, if it exposes an MCP server. Ask your SHERLOX contact for the integration form: you describe the tool, its address and its authentication method. The SHERLOX team reviews the request, then the tool becomes available in your workspace.

Yes. SHERLOX integrates with your Office environment to assist you directly where you work, without leaving your email or your document.

Yes. The SHERLOX app is available on iPhone and iPad (App Store) and on Android (Google Play): find your conversations and call on your assistant wherever you are.

To install it, search for “SHERLOX” in your phone's store, or open your account menu on the web, under “Apps and integrations”, and scan the store's QR code with your phone's camera. Then sign in with the same account as on the web.

Yes, for Windows. The SHERLOX desktop app gives you your conversations and your assistant in a dedicated window, without going through the browser, and updates itself.

To download it, open your account menu on the web, under “Apps and integrations”, and choose Windows. Run the installer, then sign in with the same account as on the web. The macOS version is in development.

Yes, in two ways. Dictation turns your voice into a message. Voice mode allows a real spoken conversation: the assistant answers you out loud, and you can interrupt it to correct or clarify.

Yes. If your request cannot be handled live (a comparison, a presentation…), the assistant confirms it is on it and the conversation continues. You are notified when the result is ready, even if you have closed voice mode, and one click takes you to it.

Yes: transcribing your voice and the assistant's voice are counted like the rest of the exchange. Anything that was not actually spoken (a noise, a silence) is neither sent nor charged.

In yours: speak in your language and it answers in the same one, with a voice that stays the same from one conversation to the next. It also recognises your company's own vocabulary.

Your credit usage is visible from your workspace: you track in real time what has been used and what you have left.

Once your credits are used up, the features that consume them are paused until your plan renews. You are warned as you approach the limit.

Yes, thanks to the private conversation: the exchange is not kept in your history nor memorized. Useful for a one-off or sensitive question.

Your credentials are encrypted and stored securely. SHERLOX uses them only to act on your behalf on the tools you have connected, and never exposes them in your conversations.

Your conversations are personal: by default only you can access them, and they are not used to train the models. If you share a conversation by link, the person who opens it gets their own copy, which they continue on their side without ever being able to write in yours. You can revoke the share.

Your company can choose to disable private conversations for all its users. In that case the option no longer appears and your exchanges are kept in your history as usual. Contact your administrator to find out more.

From your support area, you find your requests and their status, and you are notified at each step until resolution.

New features and improvements are published in the dedicated announcements / release notes area, accessible from the application.

Compaction — called “Extract context” in the interface — asks the assistant to summarize the current conversation, then opens a new thread preloaded with that summary. You start from a clean page, but the assistant still knows what you were talking about.

The longer a conversation gets, the more history the assistant has to re-read with every message: answers slow down, and it may start to confuse or forget older items. Compaction flattens the thread: only the essentials remain (goals, key information, decisions, open points) instead of the entire exchange.

The right moments:

- the conversation is getting long and the assistant answers less well or loses the thread;

- you have finished a step and are moving on to the next part of the same topic;

- you want to keep a working context without dragging along every detail of the previous attempts.

Conversely, if you are switching topic entirely, open an empty new conversation instead: carrying the old context over would add nothing.

Two paths, identical in effect:

- the Tools menu (gear icon, to the right of the input field) → “Extract context”;

- the /compact command in the input area (the command menu offers it too).

1. The conversation is turned into a transcript (“User: … / Assistant: …”).

2. The assistant receives a summary instruction: a dense, structured summary, 250 words maximum, with no greeting and no copying of the conversation.

3. A new thread is created with the same title, and the summary is stored in it as the starting context.

4. You are switched automatically to that new thread, with the summary shown in an expandable box “Context carried over from the previous conversation”.

A waiting screen is displayed during the operation: it may take a few moments, since it calls on the model.

No. It stays intact in your history, with all its messages and all its attachments. Compaction adds a thread, it deletes none. So you can always go back to the original to find a detail the summary did not keep.

No, and that is the point: it is a summary, not a copy.

- Kept: the goals, the key information, the decisions and the still-open points, as the model judged them important.

- Lost: the detail of the messages, the exact wording, and the attachments — documents sent in the old thread are not attached to the new one. If a document is still needed, upload it again.

Yes. At the top of the new thread, the box “Context carried over from the previous conversation” shows the full summary; just expand it. That is the right reflex straight after a compaction: if an important piece of information is missing, provide it again in your first message.

Two safeguards:

- the transcript sent to the model is capped: on a very long conversation, the first message and the most recent exchanges are kept, and the middle is replaced by an omission marker;

- the summary itself is capped in length before being saved.

Yes, but with one important nuance: compaction summarizes the messages of the current thread, not the summary inherited from the previous thread. On a compacted thread where you have exchanged only two or three messages, compacting again therefore does not carry the older context over. Better to compact threads that have really lived.

The cases where the button is missing or inactive:

- the conversation is in incognito mode (nothing is kept, so there is nothing to compact);

- no message has been exchanged yet;

- an answer is being generated, or a compaction is already running.

Compaction goes through the same agent as your usual exchanges: a network incident or a momentary unavailability of the model is enough to make it fail. Try again in a moment. Your original conversation is never altered by a failure. If it persists, report it to support: it is an agent configuration problem, not a problem with your conversation.

- Compact: new thread + summary of the old one. To carry on the same work, lighter.

- New conversation: blank thread, no context. For an unrelated topic.

- Share: gives someone else access to the conversation; on first opening, that person gets their own continuable copy — a mechanism distinct from compaction.