Architecture
Where it spreads, how to build it- System ArchitectLead · impact
- Software ArchitectSpec and plan
Works in Microsoft Teams and Slack
Your company,
Pick a team on the left
What Alp used
Your company,
AI-native by Monday.
@Alp joins the chats your teams already use. Every team gets its own AI specialists, and everyone sees how the work gets done.
Alp Cloud. You pay for the work your agents do.
Each team asks in its own words. Alp knows which data, which tools and which agent each question needs, and works with the asker's own permissions.
Alp checks
its own numbers.
An answer that sounds right is not the same as an answer that is right. So every morning Alp goes back to the source and checks what its agents said the day before.
- “One task is waiting in the Test column.”Jira says six. Flagged with the right number.
- “Search API: 12 of 12 pods healthy.”The cluster reports 9 of 12. Flagged with the right number.
- “Network C brought 402 leads last week.”Matches the warehouse.
- “12% of work came back from QA.”Matches Jira.
Recomputed from the source
A job with no AI in it takes yesterday's numbers and runs them again against Jira, the warehouse and your monitoring. A wrong number is flagged with the right one.
The method comes with the answer
Each data answer names the table and the field it used. Two answers that agree because they read the same wrong field get caught.
Tested like a new hire, every night
Live questions, replays of past incidents and drills where the right move is to stop and ask. A change to an agent that lowers its score is rolled back.
Counts, not people
Personal data never goes into a chat or a Jira task. Alp answers with counts and summaries instead.
From one message
to work that is done.
- 01
Ask in the chat
Anyone mentions @Alp. Dora, Alp's chief of staff, reads the thread and checks what is already in flight before she answers.
- 02
Approve the plan
Small work in one system starts right away. Bigger work comes back as a plan: what changes, where, and the risk. Reply approve, or say what to change.
- 03
Agents do the work
The right squad takes it. Each piece runs on its owner's agent, in parallel, and the Jira task moves with the work.
- 04
Checked before you see it
Two independent reviews, a live preview and a QA agent that tries to break it. Anything that fails goes back with the finding.
- 05
You decide
Going live stays with people. Alp tells the chat when the work is ready, and the whole trail stays on the issue.
Agents work in squads,
like your teams do.
A squad has a lead and members. Work goes to the lead, the lead hands each step to the member who fits it, and steps back. People can be members too.
DoraChief of staff · takes every request and runs it end to end
Delivery
Builds it, one repository at a time- Delivery LeadLead · splits the work
- Backend EngineerServices, APIs, jobs
- Frontend EngineerWeb and app screens
Review
Tries to break it before you see it- ReviewerLead · code review
- QA BreakerTests on the preview
- Security ReviewerWhen it touches security
Your squad
Any team can build one- Marketing analystLead · reads the question
- SEO analystSearch Console, rankings
- Nora KimPerson · approves campaigns
Dora takes the request
- The lead routes, it does not do the work. One hand-off per step, recorded on the issue.
- Members can be agents or people. A person in a squad gets the same hand-off an agent does.
- Squads never widen access. An agent only runs for people who are allowed to run it.
Close your laptop.
The agents keep going.
Each agent has a runtime: where its work runs. Pick it per agent, and let the work move off your laptop when you go home.
Your machine
Agents run next to your code, with your git access and your own model sign-ins.
Twin, your own cloud computer
Docker, SSH and a terminal. It sleeps when idle, keeps its disk, and takes work over from your laptop.
A pod that wakes for the work
Starts when work arrives and stops after a quiet spell, keeping your files. Sign in to Claude Code or Codex from Alp.
Workbench
Every agent session on every machine, in one browser tab.
- Worktrees on your laptop, your twin and your pods in one list.
- Open Claude Code, Codex or a terminal on any of them.
- Search every past agent session by worktree, repository or all.
- Share a running preview with a link that works for 12 hours.
- Only you reach your machines, and only after you turn it on.
Works with the tools
you already use.
Each person connects their own accounts once. Alp uses them with that person's permissions, and only for the work they asked for.
Chat
Microsoft Teams · Slack · Jira comments
Work
Jira · Bitbucket · GitHub · GitLab · Jenkins · Argo CD
Data
Google Analytics · Search Console · Google Ads · BigQuery · Postgres · Elasticsearch · Web vitals · NPS surveys
Monitoring
New Relic · Datadog · Instana · Grafana · Kibana · Sentry · Cloudflare · Uptime monitors
Agents
Claude Code · Codex · Cursor · Gemini · Grok · Copilot
Apps
Notion · Google Drive · Gmail · HubSpot · Salesforce · Linear
Hundreds more apps with one sign-in each.
An AI team that asks
before it acts.
Only the chats you approve
Alp reads a group chat only after an admin approves it, and acts only on messages that mention it.
Your permissions, never more
Alp works with the access of the person who asked. Who may start work or approve plans is set per person and per group.
People approve what matters
Plans for bigger work, risky changes and going live wait for a person. Alp never merges on its own.
A stop button and a budget
Pause all agents at once, and set daily limits on spend and runs.
Every step on the record
The request, the plan, each run, its cost and the result stay on the issue for anyone who comes later.
Limits per person and chat
Each person and each chat has an hourly cap on the work it can start, so one busy thread cannot use up the day.
Priced on the work,
not the seats.
Alp Cloud
- Everyone in your approved chats can ask Alp.
- You pay for what your agents use.
- Every run shows its cost; daily limits keep spend in check.
- We set it up with you, chats and tools included.
Questions
What is Alp?
An AI team that works in your company chat. You ask in Teams or Slack; Alp answers from your data, opens and tracks the work in Jira, and puts specialist agents on it.
Do we need engineers to use it?
No. Marketing, support, sales and operations teams ask in plain language. Engineering work goes through the same chat, with plans and reviews built in.
Which chat apps does it work in?
Microsoft Teams and Slack, and comments on Jira tasks. It replies in the same thread it was asked in.
Who can see what?
Alp works with the permissions of the person who asked, reads only approved chats, and never posts personal data to a chat; it gives counts and summaries instead.
What happens when Alp gets something wrong?
Work is reviewed and tested before anyone sees it, people approve plans and going live, and Alp rechecks its own numbers against the source every morning. You can pause every agent at once.
How is it priced?
On usage, on Alp Cloud. Talk to us and we will size it with you.
Put Alp
in your chat.
Tell us about your team. We will set Alp up with you, chats and tools included.