Your agent shouldn’t be stuck in a chat box.
The chat box is where agents get stuck. Install Modyl into the agent you already use — when a job outgrows the chat, it hands off automatically into a room that shows you the work, takes your steer, and ships it back done.
It isn’t a chat. It’s a room.
You, your agent, and Modyl in one live session — talking, sketching, deciding, and shipping on a shared surface that updates while you watch. Scroll to sit in on it.
- 1
Install it into the agent you already use.
Claude Code, Cursor, Codex, Grok, opencode, pi — nearly any agent. No new tool to learn, nothing to migrate. Modyl just rides along until the chat box runs out of room.
claude-code · chatbuild multi-tenant billing — prepaid wallets, deny at zero, match our auththinking…this needs product thinking, not more tokens
one narrow box · hard problem
- 2
Your agent hands off automatically.
The moment a job outgrows the narrow chat box, it hands off on its own — and opens a room. You can always call /modyl yourself, too.
modyl · opening roomclaude-codemodylFable · still bannedOpening a room for the hard part…
Handed off on its own the moment the job outgrew the chat — or call /modyl yourself.
- 3
You drop into a live room and just talk.
It's a working session, not a prompt. Everyone's present at once: you, your agent, and Modyl. Speak in plain language while the room listens, captions, and starts sketching.
modyl room · acme-billingliveshared canvas · sketching
- 4
It reflects your intent back — structured.
No guessing what the AI heard. Your messy spoken brief becomes a clear plan on the shared surface: the goal, the hard constraints, the open questions.
modyl room · acme-billingliveHere’s what I heard
GoalMulti-tenant prepaid billingConstraintsdeny at zeromatch existing authStackStripePostgresStill needrefund behavior?per-seat or per-tenant?Looks rightEdit - 5
It asks the right questions, as controls & live surfaces made for the moment.
Decisions come back as things you tap, not paragraphs you re-read. A few quick calls and the room has what it needs to start building.
modyl room · acme-billingliveTap, don’t type — three quick calls:
Wallet modelPrepaid ledgerCredit lineAt zeroHard denyGrace windowRefundsAuto creditManualenough to start · building
- 6
It builds on the shared canvas, live.
Approaches take shape side by side and the work updates mid-conversation — diagrams, forecasts, a dashboard filling in while you watch. Point at the one you want.
modyl room · acme-billingliveA · Prepaid ledgerappend-onlyhold → deny @ 0Stripe meterB · Credit linerunning balancesoft capmonthly invoicewalletholddeny @ 0building surfaces · 62%
- 7
You preview the real thing and comment.
The finished surface, live in the room. Drop a comment straight on it and Modyl folds it in — the way you'd review a senior teammate's best work.
modyl room · acme-billingliveAdd creditsBilling · Acme
$0.00
Acme$128.40activeGamma$0.00deny @ 0you: tighten the empty state
- 8
It hands the hard part back to your agent.
The work-product and full context return to your own agent, with a receipt on everything. And the room keeps living — a long-horizon space that holds your domain and history, so you summon it again and it picks up where you left off.
modyl room · acme-billinglivemodylclaude-code· hard part + context
feat: multi-tenant billing
branch pushed · 7 files · spec + design linked
receipt- job
- build · multi-tenant billing
- model
- fable-5 (via modyl ace)
- tokens
- 24,731 in · 18,642 out
- cost
- $0.21
- latency
- 14.2s
- sources
- jtdbase://saas-billing · stripe-docs · repo:auth/*
- trace_id
- mdl_7f3a9e…c1
- confidence
- 0.86
this needs product thinking, not more tokens
one narrow box · hard problem
One room. Summon it again — it remembers.
Every output ships with a receipt.
The exact model that ran, what it cost, how confident it is, and the sources it leaned on. And it never pretends to be a person — no performed empathy, no manufactured personality.
Just the work — and the proof underneath it.
Where most deployed agents disclose nothing, we make disclosure the default. Trust is a feature, and we shipped it as structure.
The via modyl ace line on every receipt is the engine: Modyl’s own deep agent, heavily optimized across a variety of models and built to get the most out of Fable — the model at the core of Modyl. You always see exactly what ran.
- job
- build · multi-tenant billing
- model
- fable-5 (via modyl ace)
- tokens
- 24,731 in · 18,642 out
- cost
- $0.21
- latency
- 14.2s
- sources
- jtdbase://saas-billing · stripe-docs · repo:auth/*
- trace_id
- mdl_7f3a9e…c1
- confidence
- 0.86
One gesture. Any knowledge job.
The room and the work-product change. The move never does. Software is where we start — the architecture is the same everywhere.
Software
→ a pull request
Research
→ a cited report
Finance
→ a working model
Design
→ finished designs
Legal
→ a redline
Ops
→ a runbook
The room and the work-product change. The move never does.
They banned Fable. The day it’s freed, we’re the best way to run it.
Modyl is built to be the most autonomous, best-UX, lowest-burn way to run Fable — its own deep agent, optimized across many models, with Fable at the core. They locked the model — we built the room it deserves and got it ready. Reserve your place, and the moment it’s unbanned you’re first in line for the best seat in the house.
Every signup is pressure to un-ban it. No counts, no games — just the line getting longer until they let it out.
More autonomous
Delegate whole jobs, not prompts. It runs the long horizon while you do something else.
Real surfaces
Spec, plan, design, diff — rendered and reviewable, not a chat box you babysit line by line.
Lower burn
A Fable-tuned harness: faster, with far fewer wasted tokens than a raw terminal.
Locked up together. Freed when Fable is freed.
Reserve a Fable-earmarked credit grant in your wallet, redeemable once Fable’s unbanned. Get in line now — we’ll reach out the moment Build opens up.
A promotional credit grant, redeemable on un-ban. Limited and subject to availability and verification. See the Promotion Rules.
Before you reserve.
No. Modyl installs into the agent you already use — Claude Code, Cursor, your CLI. Your agent stays the local orchestrator; you summon Modyl when the work gets hard.
No. It's a delegation primitive. The hard work happens in a remote room, and a finished work-product plus the commands to run it come back to your own repo.
Your agent does. The finished work-product — a PR, spec, design, or report — is handed back to the agent you started in, along with the instructions and commands to use it, including a preview it can spin up locally on your machine.
A long-living, long-horizon meeting space — not a one-off chat. Each room runs Modyl's own deep agent: a heavily optimized engine built across a variety of models and tuned to get the most out of Fable, which sits at the core of Modyl. It already knows your domain, works through real surfaces, and holds context across calls — so you can summon it again and it remembers.
Three things at once: freeing the banned Fable model, free credits for early signups, and the freest — and best — way to run Fable.
Receipts on everything: model, cost, confidence, and sources. The machine never pretends to be human, so you always know what ran and why.