Instead of Notion
Docs, and collections
A real block editor, and collections with 17 column types and 7 views.
Notion cannot see your production.
Replace Notion, Linear and your monitoring dashboard.
One native app, and a coding agent that reaches all three through 73 MCP tools.
macOS · Windows · Android
Your coding agent works on the same data.
- search_alertsstatus="open"
- create_taskfrom="alert"
- update_doc"runbook/api-500"
- run_in_terminal"just deploy"
Your agent knows your code. It knows nothing else.
It reads the repo: your modules, your conventions, your tests. Ask it why that error showed up, who reported it and what you had decided. It knows nothing.
The spec lives in Notion, the ticket in Linear, the CI run in GitHub, the deploy in Railway, the error in Sentry. Three families of tools, three data models, and nothing tying this morning's error to a decision made three weeks ago.
So you do the join by hand. You open the tabs, copy the error, paste the ticket, summarise the spec, paste it all back into the conversation. You have become the integration layer between your tools and your agent, and you rewrite it with every question.
What it sees
The repo.
What it cannot see
- The spec
- The ticket
- The CI run
- The deploy
- The error
- Uptime
Between the two columns, there is you, and your clipboard.
Sources
44
alert sources
Uptime is built in: Matcha pings your URLs, opens the alert, and closes it on recovery.
- Deduplicated and correlated across sources
- Full lifecycle: open, acknowledged, resolved
- Escalation, SLA and alert rules
What it replaces
Three subscriptions, one app.
Not a bridge between tools, and not a dashboard reading them from afar. Docs, tickets and alerts are native here, in one database.
Instead of Linear or Jira
A board that carries its context
Priorities, dependencies, dates, and a full page behind any ticket. An alert rule can open it for you, with its priority and project prefilled.
Linear cannot hold the runbook, or the alert behind the ticket.
Instead of a monitoring dashboard
One feed for every red light
44 sources, deduplicated on arrival and correlated across tools. Rules re-route, mute, or open a ticket.
A dashboard cannot hold the board and the docs next to the incident.
Each one is good at its own part. Wiring them together with webhooks gives you databases that send each other messages, not one database where an agent correlates an error, a deploy, a ticket and a spec in a single query. None hands a coding agent a single set of tools covering all three.
From a red light to a closed ticket, without leaving the app.
Point your tools at it
GitHub, Railway, Sentry, 44 sources in all, or a plain webhook. Duplicates collapse on arrival.
Work where the context already is
The alert becomes a ticket, the ticket carries its own page, the page links back to both.
Hand it to your agent
Your agent gets the same 73 tools your team uses, scoped to your company.
Inside
All of it exists today.
The first group comes with every seat. The other two need something on your own machine.
Day to day
Everyone · nothing to install
Alerts
44 sources in one feed, deduplicated and correlated. Rules, escalation, SLA, daily digest, a public status page, and uptime checks that open and close the alert.
Tasks
Priorities, projects, assignment, checklists, dependencies, dates. Any ticket opens a full page, and points at the alert behind it.
Docs
Versions, backlinks, templates, per-page permissions, live co-editing, and a Notion importer. Mermaid is drawn by the app itself.
Collections
17 column types, 7 views. A formula is checked server-side for syntax, references and cycles, then recomputed on display.
Terminals
Real shells in the app, driven by hand or by an agent. Every command is a block, and sharing one redacts the secrets.
Scheduled
A schedule drops a prompt into one of your agents' inboxes. Nothing is launched: the agent picks it up when it next wakes, so a closed laptop loses nothing.
If you ship a Flutter app
A module · installed per company
App flows
Replay your app's journeys with Maestro and capture every screen, run after run. Maestro runs on your machine.
Videos
The team's library: promos, walkthroughs, App Previews. Rendering runs on one of your Macs, not on a server you pay for.
What runs on your own machine
A tool or a model you bring
Agents
A crew of Claude Code sessions on your own machine, with a confined identity and a human checkpoint before anything that matters.
Meetings
Record, transcribe and summarise against the AI backend you point at. The summary lands as a Docs page.
Your own models
Point at a model and Matcha explains an alert in a few lines. Point at an embeddings model and Docs search goes hybrid.
Native core
One native library, three jobs.
Compiled into the app. No sidecar, no server, no network call.
- alacritty_terminalThe terminal
- Ours, built on Alacritty's VT core. Full-screen tools like vim and htop run properly, and your agent types in it.
- yrsPages written together
- Merges keystrokes character by character: two people in one paragraph keep both sentences.
- mermaid-rs-rendererThe diagrams
- Drawn by the app itself, with no Node and no service on the side.
Command blocks need zsh or PowerShell. Under bash the terminal works, without the blocks.
MCP
73 tools your agent already speaks
Streamable HTTP, the same authentication as the API, scoped to your company.
Alerts
- search_alerts
- get_alert
- find_correlated
- set_alert_status
- create_ticket_from_alert
- create_alert_rule
Board
- list_tasks
- create_task
- update_task
- ensure_task_page
- set_task_dependency
- add_task_comment
Docs
- search_docs
- get_doc
- create_doc
- update_doc
- query_doc_database
- update_doc_database_cell
Agents & ops
- open_terminal
- run_in_terminal
- read_terminal
- create_agent
- request_approval
- run_cron
Used by
Four teams work in it every day.
Matcha began as Despii's internal ops hub. Ganeden builds and runs it; these four teams live in it.
- DespiiWhere it started
- miissionClient
- AvanssClient
- MuwpayClient
Your agents run on your machine. Not on our servers.
The backend hands an agent its instructions, its keys and the tools it may use, then stops. The process starts on your desk.
Runs on your machine
- The agent: a real Claude Code session
- Its workspace, your terminal, your network
- The call to the model
Stays in Matcha's backend
- Its instructions and the tools it may use
- Alerts, tickets, documents: the shared record
- Its keys, minted fresh at every start
Point it at your own model
An agent can run against a model you host, as long as the endpoint speaks the Anthropic Messages API Claude Code already uses.
Inference stays on your account
The agent signs in with your own Claude Code account, so inference stays on your plan. We charge per seat, never for tokens.
Pricing
Per seat, with the ceilings written down.
Prices exclude tax. Billing opens at public launch; early access is free until then.
Free
$0
one seat
- Alerts, board, docs and MCP
- 3 sources, one company
- One agent
- No uptime checks
Pro
$12
per seat / month
- Everything in Free, up to 5 seats
- 20 sources, 25 uptime checks
- 10 agents and scheduled runs
- Escalation, SLA, daily digest
- Flutter module included
Most complete
Team
$20
per seat / month
- Everything in Pro, up to 25 seats
- Unlimited companies
- 100 sources, 100 checks, 50 agents
- Scoped API keys, one company each
- Isolation enforced in the database
Enterprise
Let's talk
custom terms
- Everything in Team, no ceilings
- Terms and invoicing to fit
Questions
Straight answers.
What is MCP, concretely?
The protocol your agent already speaks. Matcha exposes 73 tools on it, same authentication as the API.
Do I need Claude Code?
For the agent crew, yes: Matcha launches yours, on your machine. The rest works without it.
What if my machine is off?
A schedule drops a message, it does not launch an agent. The inbox is the queue: the run waits and happens once.
Does it replace my metrics dashboard?
For alerts, yes. For metrics, no: Matcha collects what your tools judged worth a wake-up.
Can I use it today?
Access opens in batches, macOS first, Windows and Android next. The invite carries the download.
Early access
Close the other tabs.
Invites go out in batches. Leave an address for the next one.
One address, one purpose. No tracker.