Observe
Present in the flow
Lives in Slack threads, DMs, channels, and side-panels instead of a detached chat tab.
AI coworkers you can run, inspect, and trust.
We build open, operator-grade AI systems that live inside the tools where work already happens. The first proof is Kortny: a self-hosted Slack teammate with memory, execution, approvals, and a visible audit trail.
Python 3.11+Docker ComposeSlack-nativeApache-2.0
01What we build
Coworker systems,
not chat windows.
The work is not a nicer prompt box. It is a runtime for teammates that can notice context, carry memory, run real work, and leave an audit trail an operator can trust.
01Observe
02Remember
03Execute
04Govern
Remember
Context with provenance
Turns work into facts, episodes, and graph edges that can be inspected later.
Execute
Artifacts, not answers
Runs tools, writes code, builds reports, and returns files or previews where the task began.
Govern
Autonomy with rails
Shows costs, traces, approvals, and tool decisions before trust has to become blind faith.
What changes
- A durable task record, not a disappearing reply.
- Memory that explains why it believes something.
- A harness that gates risky actions outside the model.
Design rule
Every claim should resolve to a visible artifact: a thread, a task trace, a memory record, a cost line, or code you can run.
02Our first coworker
Product proof
Kortny is the first coworker from the lab.
It is open-source, Slack-native, and built for teams that want a real worker without a black box. Mention it in a thread; Kortny plans, calls tools, runs code in a sandbox, remembers context, and posts the finished artifact back where the work started.
#revenue
A
Aneesh 9:41 AM
@kortny prep me for tomorrow's Acme renewal call. Pull the deal, summarize our last call, and tell me where their blockers stand.
K
Kortny orchestrating 5 tools
- HubSpotPull the Acme deal record
- ZoomSummarize last call transcript
- LinearCheck the blocking issues
- GitHubVerify the fix landed
- Google CalendarAttach the brief to the call
Posted Acme renewal brief to the thread: deal health, call recap, and the one blocker still open before tomorrow.
LinearHubSpotZoomGitHubCalendar+100 via Composio
$0.0000sandboxed code executiontask timelinesschedulesMCP servers100+ integrationsworkspace memoryknowledge graphapproval reactionssandboxed code executiontask timelinesschedulesMCP servers100+ integrationsworkspace memoryknowledge graphapproval reactions
Slack-native
Lives in the threads where your team already works.
01Self-hosted
No Matangi cloud in the path. You choose the model, integration, and deploy providers.
02Real memory
Remembers facts, past tasks, and context across time, not just one message.
03Knowledge graph
Builds a living map of your workspace: people, projects, and decisions.
04Ambient awareness
Quietly observes channels and surfaces what needs doing.
05Governed by you
Approval gates, task budgets, and traces make autonomy inspectable.
06Skills
Curated and bring-your-own playbooks for reports, research, decks, charts, and more.
07Cost visible
Per-task model, token, and cost accounting in the operator dashboard.
0803Principles
The lab standard
01
Run close to the work.
The useful place for an AI coworker is inside the team's existing workflow, with state the operator owns.
02
Autonomy needs evidence.
Every action should leave a trail: what was seen, which tool was called, what it cost, and who approved it.
03
Memory must be inspectable.
Long-term context is only valuable when a team can see provenance, scope, confidence, and stale beliefs.
04
Open systems compound.
Read the code, run it yourself, shape where it goes. Apache-2.0, all the way down.
04The name
noun · Sanskrit · मातंगी
Matangi
/məˈtɑːŋɡiː/
The goddess of speech, music, knowledge, arts and learning.
Software that listens well, holds context, and responds in language that helps.
Build with us
Build AI coworkers with proof, not mystique.
Start with Kortny, study the code, or talk to us about the next system that should live where your team already works.