Modyl is where your agents bring the calls they can't make.
Agents stop on calls a person has to make, such as architecture, scope and risk. The question arrives as a text prompt. A busy engineer clicks yes without the evidence, and two weeks later the team argues about a decision nobody really made.
- Works today
Your agent calls you into a room. Agents connect to Modyl over MCP and raise a hand when they need a decision. The room shows the question, the evidence and the options on one screen.
- Works today
Only the owner decides. Teammates mark the work. Only the engineer who owns the call can adopt it, reject it with a reason or defer it. Each answer is tied to the version they saw. Security and data
- Works today
Owners settle, curators only propose. The tool that tidies the record can only propose changes, and the database refuses anything else.
- Works today
The team keeps the record. Who decided, on what evidence, and what was turned down, for every version. The team can export it.
- Planned
Next, planned for design-partner trials. A check at plan time that holds connected agents on supported stacks to the team's settled decisions before they write the code.
What a room looks like. Finance corrects a usage number in a vendor renewal. Modyl recomputes the price and checks it, and the owner signs off.
FilmReal product on staged data. The voice and the other people are simulated. See four rooms
How it works
Intent in. Outcome out.
01
Work out the goal together
Say what you need, out loud or in writing, and bring in anyone who should weigh in. Modyl works out with you what done looks like and how to check it.
02
It does and checks the work
Modyl checks what a machine can check before it asks you anything. The agent that did the work never marks it done. Everyone you share the room with sees the work as it moves and can add input.
03
You make the call
Modyl brings you, and anyone you invite, into one room: a shared screen with the work, the proof behind it, and the call to make. Adopt the work, reject it with a reason, or defer it. A deferred call comes back on its own.
Watch it work
Four rooms. The real product.
A room holds one decision: the work, the proof behind it, and who answered. Each film is the real product on staged data. Pick a question, then watch it in the interactive room or in classic chat.
A tester asks whether the room is ready. Modyl shows the proof that passed and the one still missing, so nobody signs off early.
FilmReal product on staged data. The voice is simulated.
Solo and teams
Works on your own. Bring people in when a call needs them.
Start alone. Modyl does the work and calls you back for the decisions. When a call needs someone else, such as a colleague, a client or whoever signs off, share the room by link. They see the same work and can mark it up, and the decision stays with its owner. As more people join and more AI does the work, goals start to collide. Modyl notices the split and takes the question to one place, the person with the authority to answer it.
On your own
- 01
You set the goal. Say it out loud or write it down, then step away.
- 02
Modyl does the work. It keeps the work moving and checks it before it asks you anything.
- 03
It calls you back. Only for what needs a person: taste, judgment, and the authority to say yes.
With a team
You
Own the decision. Adopt the work, reject it with a reason, or defer it.
Today
People you invite
Open the room by link, read the work and mark it. You admit each one as they join. Only you can answer.
Today
Modyl
Proposes from what it saw and cites it. It cannot adopt its own proposal.
Today
Your organization
Sets goals, priorities and room access by policy, rooted in your corporate policy and the systems that already run it.
Nov 2026
People, teams and AI
Catch the split before the concrete sets.
Nobody hires a hundred people on Monday and expects them all to pull the same way by Friday. People need a shared goal and someone checking the work. AI workers need the same, and a team can add a hundred of them in an afternoon.
AI agents, software that carries out work on its own, now do much of the work, many at once, and nobody reads every step. When people’s goals pull apart, or a group of agents drifts off what was decided, nobody sees it until it costs days, money and people’s hours.
Agents also get in their own way. They keep acting on what they read an hour ago after the facts have changed, and that is hard to catch by hand. Modyl spots it, between people, agents or both, and keeps agents aligned before they waste time and money. When a call is needed, it puts the split in front of the person who decides while it is still cheap to fix. The person decides. Modyl never does.
FilmReal product on staged data. The voice and the call are simulated.
Who it's for
Find your work.
Each guide follows one job from the moment it starts to the call a person makes, and shows what Modyl prepares, checks and watches on the way.
- Investment committees and deal teamsTake a deal to committee with the case against still visible, and know when the thesis breaks after the vote.
- Multifamily lending teamsReconcile what was said about a deal with what the documents show, and carry every unresolved condition to closing.
- Property and asset managersExplain the month's variance, and see the corrective work through to evidence that it is done.
- Research analysts and their compliance officersKnow whether an expert call settled the question it was booked for, not only that it was allowed.
- Advisors whose client makes the decisionShow what you recommended, what your principal was shown, and what your principal chose.
- AML investigators and financial crime teamsClose or escalate an alert with a file that holds up when an examiner samples it.
- Risk owners reviewing vendorsDecide what an outside company may hold, and on what conditions, with the proof for it in the room.
- Software engineersReview an engineering question without losing its evidence.
- Product engineersReview a product question without flattening the disagreement.
The product
Keep your judgment.
See what supports a recommendation, what is still unresolved, and where your judgment is needed. Missing proof stays visible instead of disappearing into a confident answer.
The desk shows every room and where it stands: ready, settled, or still waiting on proof.
Admission is human. There is no public self-serve and no checkout. Request access and a person reads it.
Desk
Rooms
- Should the human tester be the evaluator of record?Not yet · one proof missingOct 22, 12:00 PM
- Vendor renewalAdopted · bridge at $140kOct 22, 12:00 PM
- Billing to the new ledgerSettled by Dana · agents resumedOct 22, 12:00 PM
The rooms from the films above, rebuilt from a real Modyl desk screen for legibility. Times follow today’s date.
Works today
Modyl does the checking. You bring the judgment.
Everything in this list works today, in invite-only early access.
For everyone
Works today- Modyl brings its own agent. No agents of your own, or rather not run them? Modyl's agent works toward the goal against its checks.
- Done means checked. For work that can be tested, an AI from a different company writes checks the working AI never sees. The agent that did the work never marks it done.
- Review in a room. Mark the work, then adopt it, reject it with a reason, or defer it. Each answer is tied to the version you saw.
- Guests mark. Only you answer. Invite people by link and admit each one as they join. Revoke the link and they are out.
- Decisions come back. Defer a call to a date, or until another decision is adopted. It returns to the room on its own.
- It moves on events, not prompts. A tripped rule, a failed check, a replaced decision or a call coming due sets Modyl going. Nobody has to ask first.
- Always on, in any medium. Voice, text, video and built visuals, with Modyl calling you when something needs you.
- Talk to it. Live voice in the room. Talk over it, correct it, and keep going.
- Connect a tool while you talk. Modyl puts a sign-in link in front of you and picks the work back up the moment you approve it.
- Any work in the room. Decks, docs, videos and apps, with every new version keeping your marks.
When goals collide
Works today- Modyl raises the split itself. It notices divergence between people, between agents, or between both, and calls whoever holds the decision.
- The right person is called in. A tripped rule goes up the chain of agents to the person who set the intent, or says plainly that it cannot find one.
If you already run agents
Works today- Bring your agents. ChatGPT, Claude, Claude Code, Codex, Amp and agents you built yourself connect over MCP or a connector.
- Your agents can call you in. Agents connect over MCP and raise a hand when they need a decision.
- Agents get the settled answer. A new decision replaces the old one, and agents read the current answer over MCP before they carry on.
- Rules that hold across sessions. A confirmed rule fires in every connected session of a workspace. In Claude Code, Codex and Amp it can stop a matching action before it runs. If the check takes longer than 2 seconds or cannot run, the call goes through. It catches some misconduct, not all.
- Drift gets spotted. Modyl notices agents drifting off the goal or stuck on stale assumptions, and steers them back before they waste time and money.
- One decision across many agents. Modyl spots agents still working from a replaced decision, or from a goal that moved, and points them to the current answer.
What’s next
Where Modyl is going.
Planned. Not available today.
Nov 2026
Planned- Goals held for weeksModyl pursues a goal against its checks and runs the research on its own, until the goal is met.
- Goals deconflicted up frontModyl compares what each person and team is pursuing, shows each side with its source, and asks whoever holds the decision before anyone builds on it.
- The cost curve, watchedSpend, repeated failures and agents handing work to more agents trigger a call before the bill grows.
- Questions shaped to the judgmentPick one, rank these, mark this region. Structured where that helps you focus. Open everywhere else.
- Knows the teamWho owns what and who decides. Goals follow your organization’s policy and the systems that already manage it.
- Goals that keep up with the workModyl keeps refining the goal, what done looks like and the checks as the work and your answers reveal more, and asks before it changes what counts as done.
- Gets better with useModyl keeps learning from every answer, rejection and reason. Your calls become what it measures its next version against, for you and for your organization.
- Zero data retention, on EnterpriseEnterprise workspaces can require zero data retention from the providers that handle their content. Until then, research providers handle queries under their standard retention terms, as our privacy policy says.
Dec 2026
Planned- Deploy where you need itEnterprise deployments in your own cloud account, on your own servers, or a mix of both.
- Bring your own AI provider keysModyl runs its AI on your own account with an AI provider.
- Bring your own keys for encryptionEncrypt workspace content with keys you hold.
You're holding it wrong.
With Modyl, there's no wrong way to hold it.
Say what you want, don't hold back. Settle what matters. Keep the record.
Modyl frees frontier AI to work for you.
Campaigns
The case against the chat box, five ways.
Start with the friction
What part of the work needs fixing most?
Name the recurring ceremony or work product that keeps slowing down. Give us as much or as little context as you want.