The sovereign, multiplayer, extensible and open CompanyOS
Hand it an outcome and AI agents run it 24/7. Connect your company’s tools, share agents and context across your whole team, and own all of it as code: open source, on your infrastructure or ours.
Type / for skills, commands, and files
- 1# yaml-language-server: $schema=https://kortix.com/schema/kortix.schema.json
- 2default_agent: kortix
- 3
- 4project:
- 5 name: arcadia
- 6 description: Arcadia runs engineering, marketing and sales on Kortix.
- 7
- 8env:
- 9 required: [GITHUB_WEBHOOK_SECRET, LEADS_WEBHOOK_SECRET]
- 10
- 11# Connectors: the credentials live in Kortix, never in this file.
- 12connectors:
- 13 - slug: github
- 14 provider: composio
- 15 app: github
- 16 - slug: linkedin
- 17 provider: composio
- 18 app: linkedin
- 19 policies:
- 20 - match: "linkedin.create_*"
- 21 action: require_approval
- 22 - slug: hubspot
- 23 provider: composio
- 24 app: hubspot
- 25
- 26skills:
- 27 triage-issue: skills/triage-issue
- 28 write-post: skills/write-post
- 29 qualify-lead: skills/qualify-lead
- 30
- 31tools:
- 32 lookup_account: tools/lookup_account.ts
- 33
- 34commands:
- 35 draft-post:
- 36 description: Draft a LinkedIn post about one change.
- 37 prompt: "Draft a LinkedIn post about $ARGUMENTS. Follow the write-post skill."
- 38
- 39# pi extensions: TypeScript that pi loads into the agent's session.
- 40extensions:
- 41 triage-policy:
- 42 pi: extensions/triage-policy.ts
- 43 web-access:
- 44 pi: npm:pi-web-access@0.30.0
- 45
- 46# Every agent is a pi agent. It gets only what its block lists.
- 47agents:
- 48 kortix:
- 49 description: The general agent for Arcadia.
- 50 mode: primary
- 51 prompt_file: agents/kortix.md
- 52 connectors: all
- 53 skills: all
- 54 extensions: all
- 55 workspace: { project_repository: true }
- 56
- 57 # Engineering
- 58 eng-triage:
- 59 description: Triages every new GitHub issue.
- 60 prompt_file: agents/engineering/triage.md
- 61 tools: [read, grep, glob, web_search]
- 62 connectors: [github, kortix_slack]
- 63 skills: [triage-issue]
- 64 extensions: [triage-policy]
- 65 workspace: { project_repository: true }
- 66
- 67 # Marketing
- 68 marketing:
- 69 description: Drafts LinkedIn posts from shipped work.
- 70 prompt_file: agents/marketing/content.md
- 71 tools: [read, web_search]
- 72 connectors: [linkedin, kortix_slack]
- 73 skills: [write-post]
- 74 commands: [draft-post]
- 75 extensions: [web-access]
- 76
- 77 # Sales
- 78 sales-inbound:
- 79 description: Qualifies every inbound lead.
- 80 prompt_file: agents/sales/inbound.md
- 81 tools: [web_search, lookup_account]
- 82 connectors: [hubspot, linkedin, kortix_email]
- 83 skills: [qualify-lead]
- 84 extensions: [web-access]
- 85
- 86triggers:
- 87 - slug: new-issue
- 88 name: Triage new issues
- 89 on: { webhook: { secret: GITHUB_WEBHOOK_SECRET } }
- 90 run:
- 91 agent: eng-triage
- 92 prompt: "Triage issue #{{ body.issue.number }}: label it, find the owner, post in #eng."
- 93 - slug: inbound-lead
- 94 name: Qualify inbound leads
- 95 on: { webhook: { secret: LEADS_WEBHOOK_SECRET } }
- 96 run:
- 97 agent: sales-inbound
- 98 prompt: "Qualify {{ body.email }}, look them up on LinkedIn and log them in HubSpot."
- 99 - slug: weekly-post
- 100 name: Weekly LinkedIn post
- 101 on: { cron: "0 0 10 * * 2", timezone: Europe/Berlin }
- 102 run:
- 103 agent: marketing
- 104 prompt: "Draft this week's LinkedIn post from merged pull requests."
A workspace and command center for people. A backend framework for agents underneath. The same project, as files in a git repo you own.
Models and apps Kortix works with
- ClaudeClaude
- SlackSlack
- OpenAIOpenAI
- NotionNotion
Multiplayer. Company as code. Sovereign. The CompanyOS you own.
Multiplayer AIOne AI for the whole team.
Everyone works with the same agents, skills, memory and connectors. Start a session in the web app, in Slack or in Microsoft Teams.
- Shared agents, skills and memory
- Slack and Microsoft Teams
- Roles, groups and per-resource permissions
- A person approves what reaches main
Company as codeExtensible to the core.
Every agent, skill, rule and connector is a file in one git repo. Each agent is a custom pi agent you extend with pi extensions.
- Every agent and skill a file in git
- Custom pi agents and pi extensions
- 3,000+ apps, plus MCP, OpenAPI and raw HTTP
- A CLI and a TypeScript SDK
Sovereign and openOwn your data and your infrastructure.
Kortix is open source. Run it on Kortix Cloud, in your VPC or on your own on-prem network, with any model and your own keys.
- Open source: read it, fork it, audit it
- Self-host in your VPC or on-prem
- Any model provider, your own keys
- Secrets encrypted per project
How Kortix works.One repo, every tool, any model, a computer per agent.
- New skill: weekly reportSkillsnow
- Finance can read billingRules2m
- Runs every Monday at 9:00Triggers1h
- Support agent addedAgents3h
Every change lands in acme-co
A workspace for your team.A framework underneath.
People work in the Kortix workspace. Underneath, Kortix is a backend framework for agents: a project is a git repository, and kortix.yaml declares its agents, skills, tools and triggers.
Every agent is a custom pi agent: the open-source pi coding agent plus your prompt, skills and pi extensions, deployed to the cloud. Every session runs on its own computer. Triggers run agent loops 24/7. @kortix/sdk and @kortix/ui let you build your own experience on the same backend.
kortix.yaml
Agents, access, tools and triggers in one versioned file.
pi coding agent
The open-source harness every agent runs on. Extend it with pi extensions.
@kortix/sdk
One typed client for projects, sessions and live events.
@kortix/ui
The React components the Kortix workspace is built from.
Outcomes
Give it an outcome. Agents run it 24/7.End to end, for every team.
An SEO and GEO program that runs itself.
It tracks the queries you care about in search and in AI answers, finds the pages losing ground and rewrites them against the brief. Then it measures what moved, learns from it, and runs the loop again every week.
You
Run our SEO and GEO loop: track our queries in search and AI answers, and rewrite the pages losing ground.
Everything IT needs to say yes.Agents for the whole team, without giving up control.
Access control
Per-resource permissions for people and agents, with roles and groups.
Guardrails
Allow, Ask or Block per tool. Agents wait for a person where you say so.
Full audit trail
Every action an agent or a person takes is recorded.
Isolated machines
Each session on its own machine. Secrets encrypted per project; connector credentials stay on the server.
Single sign-on
SAML 2.0 single sign-on and SCIM 2.0 directory sync.
No lock-in
Any model with your own keys, on Kortix Cloud, your VPC or on-prem.
- Security at Kortix
SOC 2 Type I verified
GDPR operated. Open source, so your team can read every line.
The questions people ask.Answered plainly.
Anything else, write to the team.
What does an agent actually run on?
Its own machine. Every session boots an isolated Linux cloud computer with your repo and your tools already on it, on its own git branch — the session id, the sandbox id and the branch name are one and the same string. The agent can install, run and break anything inside it; only what it commits survives, because the machine is destroyed with everything on it.
What stops it doing something I did not want?
You set allow, ask or block per action, and a rule can match on a glob, a regular expression, or the arguments of the call itself. Be aware that nothing is switched on for you: a project with no policy declared runs every action, so this is a thing you configure rather than a thing you inherit. Session work reaches main through a change request, and merge is default-deny for an agent unless an admin grants it.
Can I see everything an agent did?
It is all recorded, on every plan. Recording is never gated — actions, approvals and the session transcript are captured whatever you pay. Reading, exporting and streaming that audit trail out to your own SIEM is an Enterprise entitlement, so on the lower plans the record exists but the export does not.
Is my company data yours?
No. A project is a git repository — clone it and the agents, the skills and everything the company has learned come with you. Self-host and the database, the files, the repos, the secrets and the audit record all sit on disk you control, on one Docker Compose stack built from the same images the managed cloud runs. It is not air-gapped, and we will not pretend otherwise: agent sandboxes run on the compute provider you configure, and your instance has to be reachable so they can call back.
Can I use the models I already pay for?
Yes. Bring your own API key for any major provider, or sign in with the ChatGPT subscription you already pay for, or use the managed gateway and bring no key at all. The model is a line of configuration in your repo, so switching provider is a diff rather than a migration.
What does it cost?
Free is $0 — one project, 500 credits a month and every feature, with your own key or ChatGPT subscription for models. Kortix-managed models need a paid plan. Paid plans are credits: pick a monthly amount, members are free and unlimited, and one pool covers the whole account. Self-hosting the software costs nothing; you pay your own model bill and your own compute bill instead.