Features and their boundaries

The problem

After a week of calls and documents, where a client project stands lives in your head and in scattered notes. Working out what changed, and what is now waiting on you, means rereading everything.

The outcome

You open a project and see what is true now, what changed and when, and which proposals or commitments need you, each pointing back to its source, so you can choose the next safe move.

What it is designed to own

• A current project state across decisions, commitments, tasks, and risks, each linked to its source.
• A history of what changed and when: a decision is designed to be superseded or reversed without losing what came before.
• Our aim is an attention queue that brings forward proposals waiting for review and commitments coming due.

What it does not own

• It is not designed as a project-management suite or an issue tracker, and your task tracker stays where it is.
• The attention queue is not meant to act for you: it is not designed to assign, send, score, or change anything.

How it would work

1. Records you have confirmed from your sources make up the project state.
2. When a later conversation changes something, the earlier version is designed to stay in the history.
3. What needs you next is meant to come forward with its receipt, and you decide what to act on.

Illustrative example: Midfield rebrand The logo review moving to next Tuesday replaces the earlier Friday date, which stays in the history, and the Thursday brand-guidelines commitment comes forward as it falls due.

Privacy and control

• Each item is designed to be traceable to its source, so you can check why it is there.
• You decide what to act on. Nothing in the queue is designed to happen without you.

Status and caveat

Status: PlannedStatus: Design direction

Project state and history are planned. The attention queue is a design direction, not a product promise. Neither is in a build you can try.

Who early access suits

Studio leads and consultants who run several client projects at once and lose time rebuilding where each one stands.

Who should not join for this

Anyone looking for a replacement project-management or issue-tracking suite, or for scores and reports on how people work.

• Step 5 of how it works→
• Step 6 of how it works→
• Do I need to replace my tools?→

Request early access

The problem

Promises made on calls slip because nobody wrote down who owes what, to whom, by when, or where it was agreed. Later, it is one memory against another.

The outcome

Each decision and commitment is kept with its owner, any agreed date, the tasks and risks it affects, and the receipt showing where it was said, so follow-through does not depend on memory.

What it is designed to own

• Decisions with their rationale and history, kept apart from commitments, tasks, and risks.
• Commitments with who owes them, who they are owed to, a due date where one was agreed, and whether they have been delivered.
• Links from each record to the tasks and risks it affects and to its source receipt.

What it does not own

• It is not designed to turn every action item into a commitment, or to assign anything to anyone without your review.
• Your task tracker and email stay where they are; Clarni is designed to sit alongside them.

How it would work

1. Suggestions from a source arrive as proposals, each with its receipt.
2. You confirm, edit, or reject each one. Extraction can be wrong.
3. Confirmed items are designed to be kept as distinct records, linked to each other and to their receipts.

Illustrative example: Midfield rebrand “I will send the final brand guidelines by Thursday” becomes a commitment owed by the Midfield brand lead to your studio, and a proposed task nobody agreed to is rejected.

Privacy and control

• Nothing is designed to enter the record, or be assigned to anyone, until you review it.
• You are meant to be able to open the receipt behind any record and check it against what was said or written.

Status and caveat

Status: Planned

Planned. Suggestions are proposals and can be wrong, so reviewing them stays with you.

Who early access suits

Consultants and studio leads who make and receive promises across many client calls and need to show where each one came from.

Who should not join for this

Anyone who wants obligations assigned or tracked without reviewing them first.

• Step 3 of how it works→
• Step 4 of how it works→
• Can I trust what the AI extracts?→

Request early access

The problem

Writing the follow-up after a call means reconstructing what was agreed, and a confident sentence nobody actually said can end up in front of a client.

The outcome

A draft built from the sources you choose, each line pointing to its receipt, that you edit and approve before deciding whether to send it.

What it is designed to own

• Drafts grounded in the sources you select, with each line linked to its receipt.
• Editable, versioned drafts kept on your device, so an earlier version is not lost when you change one.

What it does not own

• The draft view is designed to send nothing, and a draft is never a record of what was agreed.
• It is not designed to replace your email, chat, or documents: a draft goes out through them only after you confirm.

How it would work

1. You choose the recipient and the records and receipts to draw on.
2. A draft is proposed with each line linked to its source, and you edit any word.
3. If you decide to send, you see what will go where and confirm immediately before it happens.

Illustrative example: Midfield rebrand Version 2 of the follow-up confirms the Tuesday logo review and asks for the guidelines by Thursday, with receipts 12:34, 08:15, and 14:02 behind its lines.

Privacy and control

• You are meant to pick the sources a draft may use before anything is generated.
• Nothing is designed to be sent, published, or written back to another system without your explicit confirmation.

Status and caveat

Status: Planned

Planned. Drafts can contain mistakes, so you are meant to check each line against its receipt.

Who early access suits

People who write many client follow-ups and status updates and want each line traceable to what was said.

Who should not join for this

Anyone looking for bulk outreach, or for messages that go out without a person reading and confirming them.

• Step 7 of how it works→
• Step 8 of how it works→
• Will it message my clients for me?→

Request early access

The problem

What was said on a call or written in a document is hard to find again when you need to check it, and you should not have to hand over your whole week to find it.

The outcome

You bring in the sources that matter to a project, then find where something was said or written and cite it next to the record that came from it.

What it is designed to own

• Imports you choose: a recording, a note, or a document, added to the project you pick.
• Searchable transcripts made on your device from the recordings you bring in.
• Source receipts with the excerpt, the speaker where known, and the timestamp or page that records and drafts point back to.

What it does not own

• It is not designed to watch your clipboard, your screen, or your communication apps. Nothing is meant to come in unless you bring it.
• Transcripts are meant as an input you can point back to, not the product, and your documents stay in the tools where you keep them.

How it would work

1. You pick a source and the project it belongs to.
2. It is kept as evidence: excerpts, speakers where known, and timestamps or pages.
3. Later, you search it and open the receipt behind any record.

Illustrative example: Midfield rebrand You bring in the weekly sync recording and the working copy of the brand guidelines, and nothing else from your week.

Privacy and control

• You are meant to choose each source and the project it belongs to.
• Bringing a source in is not meant to send it anywhere: it is designed to stay on your device unless you later choose a cloud AI action.

Status and caveat

Status: Planned

Planned. Transcripts can contain errors, so each receipt keeps the original excerpt to check against.

Who early access suits

People who need to find and cite what a client actually said or wrote, without collecting everything they work on.

Who should not join for this

Anyone looking for a stand-alone notetaker, or for calls and messages to be captured without choosing them.

• Step 1 of how it works→
• Step 2 of how it works→
• Is this just another AI notetaker?→

Request early access

The problem

Client work is sensitive, and many AI tools send it to a provider before you have seen what is going, or to whom.

The outcome

AI help that starts on your device and reaches a cloud provider you chose only when you have opted in and trigger an action, with the selected context shown first.

What it is designed to own

• Work data stored on your own device by default, encrypted.
• A local model as the default route for AI actions.
• Optional cloud modes you switch on yourself, with a choice of provider and a view of the selected sources and the provider before anything is submitted.

What it does not own

• It is not designed to choose a cloud provider for you, or to switch one on without your explicit choice.
• A cloud provider you opt in to handles what you send it under its own terms; Clarni does not set them.

How it would work

1. An AI action you trigger is designed to run on a local model by default.
2. If you have opted in to a cloud provider, you see which sources and which provider before anything is submitted.
3. You submit or cancel, and the result comes back as a proposal or draft for you to review.

Illustrative example: Midfield rebrand Drafting the follow-up on a local model is designed to keep the receipts on your device. With a cloud provider switched on, you would see the selected receipts and the provider first.

Privacy and control

• Local-first is not offline-only: work data is designed to leave your device only when you choose a cloud AI provider and trigger an action, or confirm an external action.
• Cloud modes are designed to stay off until you switch them on, and to send only the context you selected.

Status and caveat

Status: Planned

Planned. Models on your own device may be slower or less capable than cloud models, and we make no speed or quality promise.

Who early access suits

People whose clients expect their material to stay under the professional’s control, and who want to decide case by case whether an AI provider sees it.

Who should not join for this

Anyone who needs a guaranteed offline-only setup, a specific model or provider, or a set speed or quality from on-device AI.

• The boundary in how it works→
• Which AI does it use?→
• Where your data lives→

Request early access

The problem

Connecting an assistant or a work tool often means granting broad access and hoping it does only what you meant.

The outcome

An assistant reads only the project context you share and suggests changes for your review, and anything bound for another system waits for your confirmation.

What it is designed to own

• Read: assistants are intended to read only the project context you choose to share, set on your device.
• Propose: changes from an assistant are designed to arrive as proposals you review like any other.
• Write: anything bound for an email, chat, or document system is designed to wait for your explicit confirmation immediately before it happens.

What it does not own

• No assistant product, integration, or provider is named on this site: support depends on each provider’s official capabilities and on tested compatibility.
• It is not designed to replace your email, chat, documents, or issue tracker, or to become an integration marketplace.

How it would work

1. You choose which project context an assistant may read.
2. The assistant suggests a change, which arrives as a proposal with its source.
3. Before anything reaches another system, you see what will go where and confirm or cancel.

Illustrative example: Midfield rebrand Sending version 2 of the follow-up by email waits on a confirmation that shows the recipient and the exact text. If you cancel, nothing is sent.

Privacy and control

• Reading, proposing, and writing are designed to stay separate: letting an assistant read does not let it write.
• No connection is meant to carry out an external action without your confirmation just before it happens.

Status and caveat

Status: Planned

Planned. Compatibility with a particular assistant or tool is not promised and depends on testing against each provider’s official capabilities.

Who early access suits

People who already work with an assistant and want it to see the right project context without handing over everything.

Who should not join for this

Anyone who needs a specific named integration, or an assistant that acts in other systems without a person confirming each action.

• Step 8 of how it works→
• Can assistants read my Clarni context?→
• Where your data lives→

Request early access