AI & Agents

Agents that do the work.
Audited every step.

Message an agent in any channel and it responds — no slash commands. Agents check on tasks and incidents, research competitors, draft notes, and route documents for approval. On iPhone, Voice AI lets you create tasks and summarize today’s messages without a keyboard. Pick the model for the job, then bring Cursor, Claude, and other tools through MCP with revocable, audited grants.

Get started freeSee pricing
No credit card · explore alone or invite your team · about 60 seconds
app.fabricloop.com
fl
H
Home
T
Tasks
N
News
Nt
Notes
Tm
Team
Ag
Agents
Ap
Apps
Home›Agents
Agents
Author automations for this workspace. Each agent needs a home channel; iOS consumes runs and questions.
Spend leaderboard (30d) — highest first.
Datadog (US1)
#channel · read_only · Balanced · GPT-5.6 Terra · $0.00 / 30d
active
Competitive Research
#Product · read_only · Thorough · GPT-5.6 Sol · $0.00 / 30d
paused
Loop Agent
#Product · read_write · Balanced · GPT-5.6 Terra · $0.00 / 30d
active
Onboarding Concierge
#Marketing · read_only · Thorough · GPT-5.6 Sol · $0.00 / 30d
paused
In channel

Just message the agent. It replies in thread.

Apps and agents are participants in the conversation. Ask in the channel where the team already works — they answer in thread. No /bot syntax, no separate command palette. @Jira, tell me what tasks I have today. @Datadog, what’s paging right now. @Competitive Intelligence, tell me about the latest model from Anthropic. @Loop Agent, summarize this thread into a note.

01
Any channel, same habit
Talk to an app or agent like a teammate — in DMs or Groups.
02
No slash commands
If people have to memorize syntax, they will not use the automation.
03
Context stays in the thread
The agent sees the conversation. Replies land where everyone can follow.
#eng
MV
Maya9:14
@Jira tell me what tasks I have today
JI
JiraApp9:14
You have 3 due today: Review Q2 pricing (5pm), Ship onboarding email, and Approve vendor contract.
#product
AD
Alex9:16
@Competitive Intelligence tell me about the latest model from Anthropic
CI
Competitive IntelligenceAgent9:16
Anthropic’s latest is Claude Opus 4.1 — stronger long-context coding than Sonnet, same API. I filed a one-pager in Notes if you want to share it.
#ops
CW
Chris9:21
@Datadog what’s paging right now?
DD
DatadogApp9:21
One active page: api-gateway latency p99 above 800ms for 12 minutes. I posted the graph in the thread.
#product
AV
Alicia9:28
@Loop Agent summarize this thread into a note
LA
Loop AgentAgent9:28
Drafted “Q2 pricing follow-ups” in Notes — three decisions, two open questions. Confirm and I’ll file the tasks.
Loop Agent

AI that does work, not just answers.

01
Ask across tasks, notes, and threads
Get answers grounded in your Groups — not a blank chat window.
02
Turn any message into work
One action turns a message into a task, note, or news — AI drafts it from the conversation so you confirm and send.
03
Create tasks and draft updates
Turn natural language into owned work and notifications.
04
Summarize calls and conversations
Close the loop so the team sees what changed.
05
Human-in-the-loop by design
Agents assist; people stay accountable for the send.
Models

Pick the model for the work.

Agents can run on Anthropic’s Haiku, Sonnet, or Opus, or OpenAI’s GPT-5.6 tiers — Sol, Terra, and Luna — chosen automatically based on the work. We bill model costs directly at provider rates.

01
Anthropic Haiku / Sonnet / Opus
A capability ladder, not just speed — Haiku is fastest and cheapest, Sonnet balances cost and capability, Opus is the most capable and most expensive.
02
OpenAI GPT-5.6 Sol / Terra / Luna
OpenAI’s official GPT-5.6 tiers — Sol is the flagship, Terra balances cost and capability, Luna is the fastest and cheapest.
03
Example costs
Summarizing a 50-message thread is typically a few cents. A light 20-person workspace is often in the $5–$20/month range. Hard spending limits live in Settings → Billing.
04
Direct model billing
Free includes test-only credits. Paid plans bill production usage as you go. Bring-your-own provider key is on the roadmap, not available today.
MCP

Client and server. Grants you can revoke.

FabricLoop ships an MCP client and its own MCP server. Connect the AI tools you already use, list grants, and revoke access — with audit events for procurement and security review.

01
Supported tools today
Cursor, Claude, and ChatGPT via MCP to read and write notes, tasks, and news. Other MCP clients can connect the same way.
02
What a grant can access
Each connection is scoped — typically notes, tasks, and news in Groups the connecting user can already see. It does not get a workspace-wide backdoor.
03
Default: human approval
Agents draft; people send. Loop Agent does not silently assign work or post news without an approval step.
04
Audited, revocable grants
See who connected what — and turn it off when the project ends. Enterprise adds SSO and broader audit logs.
Voice AI

Talk to it on iPhone. It does the work.

Voice AI on iOS lets you actually create tasks and summarize your messages for today — while you are on your iPhone, not back at a desk. Free includes 2 voice requests per hour; paid plans lift the cap.

01
Create a task by speaking
Capture owned work from the sidewalk or between meetings — without opening a keyboard.
02
Summarize today’s messages
Ask what you missed this morning. Voice AI reads your channels and tells you what actually needs a reply.
03
Native iPhone, not a desktop leftover
Built into the iOS app. Hands-free when typing is slower or you are away from a laptop.
04
You still confirm
Voice drafts the task or the summary. You approve before anything is assigned or posted.

New. Better. Native. Fast.

Messaging with work attached. Connect Cursor, Claude, or ChatGPT via MCP to read and write notes, tasks, and news. Installable apps on every platform.