The governed agent operating system.
Everything your AI agents need to do real work, in one place: identity, memory, tools and any model, under rules you set.
- A Co-worker for every employee
- AI teammates for every team, right in Slack
- Claude, ChatGPT and your own agents, on the same rules
Priya asks her Co-worker in Slack to update a deal in HubSpot.
Priya messages her Co-worker where she already works.
Claude, ChatGPT, GitHub Copilot and your own agents through one secure gateway.
Files and memory that carry across every agent and session.
Co-workers in Slack, meetings and scheduled routines.
Policy, approvals and audit on every tool call, before it runs.
Keep the assistants your teams already use. Claude, ChatGPT, GitHub Copilot and any MCP client connect through the same governed gateway, and every tool call they make follows one policy.
ClaudeInteractiveClaude DesktopInteractive
ChatGPTInteractive
GitHub CopilotInteractive
VS CodeInteractive
GooseInteractive
PostmanInteractive
MCPJamInteractive
Archestra.AIInteractive
CursorInteractive
Windsurf
Zed
JetBrains
Cline
Continue
Gemini
LangChain
LangGraph
CrewAI
Vercel AI SDK
Python SDKChatGPT and Claude are great assistants. Your company needs the layer underneath.
Your teams will use more than one AI. Each assistant governs only itself, remembers only what happened inside it and runs the models its vendor offers. CortexHub is the layer every agent shares.
Keep the assistants your teams like. Connect them through CortexHub and they work with the same memory, tools and rules as everything else.
Every agent your company runs, on one governed platform.
Give every employee a Co-worker, give every team a shared agent, bring the assistants people already use and govern the agents in your own product. Every tool call they make goes through the same gateway, the same policy and the same audit trail.
Governed Co-workers
A personal AI co-worker in web, mobile, Slack and meetings that builds real deliverables and runs routines.
Meet the Co-worker For every teamOrganisation agents
Shared AI teammates owned by the organisation, on company credentials, with their own name in Slack.
See team agents For the assistants you already useAssistants
Claude, ChatGPT, GitHub Copilot and any MCP client connect over MCP. Every tool call they make follows the same policy.
How the gateway works For the agents you buildProduct agents
Agents in your own product, on any framework, connect through the SDK, API or MCP with one scoped key per agent.
For developers- 1IdentityWhich agent, acting for whom
- 2PolicyAllowed for this agent, tool and data?
- 3Content safetyPrompt injection and sensitive data
- 4ApprovalA person decides when the rule says so
- 5AuditTamper-evident record of the decision
GitHubRead pull requestAllowedGitHubMerge to mainNeeds approval
SlackPost to #customersNeeds approval
NotionUpdate project pageAllowed
SalesforceDelete accountBlocked
StripeIssue large refundNeeds approvalYour context goes wherever your agents go.
Files, memory and work in progress live in CortexHub, not in any single AI. Start in Claude, continue in ChatGPT, hand off to a co-worker in Slack. Nothing is lost and nothing starts over.
Claude
via the CortexHub MCP gateway
Drafts the Q4 pricing one-pager, saves it to your files and remembers the enterprise tier decision.
ChatGPT
same workspace, different assistant
Recalls Monday's decision, opens the same file and saves a tighter version 2.
Ava, your co-worker
in Slack, on a schedule
Turns version 2 into a deck for #leadership. Sending it to a customer waits for your approval.
One Brain across your agents
Decisions, people and projects your agents learn stay in CortexHub, not inside one assistant.
Files any of your agents can open and edit
Any format. Every save is a new version, whoever made it, with the latest one on top.
Governed every step of the way
Every read, write and send goes through the same policy, approvals and audit trail.
One ask becomes a dashboard the whole team can use.
A Co-worker pulls live data, builds an interactive dashboard and saves it to Files. Share it with a colleague and their own Co-worker can work with it at once. Nothing is copied: one file, one set of versions, and access ends the moment you unshare.
Stripe
HubSpotTry the Q2 filter Priya and her Co-workerWork that runs itself, step by step, with the reasons shown.
Describe the job once and put it on a schedule. Every run shows each step, why the agent took it, which model it used, what it cost and where a person approved. From a five-person team to a regulated enterprise.
- 1Find invoices unpaid for 14+ daysRunning
Why: Monday is chase day; the owner only wants invoices past their terms.
Stripe6 invoices, 8,420 EUR - 2Draft a polite reminder for each customerDone
Why: Customers who paid late before get a firmer tone, per last month's note in the Brain.
- 3Send the 6 remindersApproved
Why: Emails to customers need the owner's sign-off under the finance policy.
GmailApproved by Sarah in Slack, 08:06 - 4Post a summary to #financeDone
Why: The team tracks cash collection in one channel.
Slack
One living map of your work, not a black box.
The Brain is a single living map of your work: the people, projects, conversations, documents, preferences, and lessons your Co-worker learns. It grows from everyday work, keeps one canonical record of each person and project, remembers what matters, and corrects itself when things change. You can see it, explore it, and correct it, so what the Co-worker knows is transparent and auditable.
- One record for each thing. Explore by People, Projects, Topics, or a graph view. "The migration" and "Q3 Migration" stay one record, not duplicates.
- See and correct it. Pin what matters, retract what is wrong, and merge duplicates. Corrections become lasting, governed knowledge.
- Governed and auditable. Every read and write runs through the same policy and audit as any other agent, so you can audit what it knows.
A Co-worker that shows up and speaks up.
Your governed Co-worker does not just take notes. It joins the call already knowing your people, projects, and decisions from the Brain, speaks in its own voice, answers questions, and contributes. When the meeting ends, what was decided flows back into the Brain, governed by the same policy as everything else.
An active participant, not a silent notetaker. It participates and remembers. It does not replace the people in the room.
- The Brain is why it is good. Because the Co-worker learns from everyday work, it walks into the meeting prepared. The better the Brain, the better the participant.
- Every meeting makes the Brain richer. What is decided in the room becomes lasting, governed knowledge, so the next conversation starts further ahead.
- Governed like everything else. The same policy, approvals, redaction, and audit trail apply in the meeting as on every other tool call. No separate rules.
“Picking up the renewal thread: last quarter we paused Project Atlas on the security review. That sign-off cleared Tuesday, so we can commit the rollout date today.”
Walks in already knowing.
Before the call, the Co-worker draws on the Brain: your people, projects, past conversations, and the decisions that led here. It arrives with context, not a blank page.
Joins the call and contributes.
It is in the room. It speaks in its own voice, answers questions, presents decks, and pulls live numbers or builds a chart while it keeps talking. It treats the call as internal only when everyone present is verified as a colleague (on Google Meet, by matching each name to one verified colleague); otherwise it holds back internal detail.
Every decision flows back.
Decisions, action items, and follow-ups land in the Brain, under the same policy and the same audit trail. Every meeting makes the Brain richer for the next one.
What happens when something goes wrong.
Four real situations, played out the way CortexHub handles them. Every guard acts in the path of the agent and leaves a record (a risk signal, or an alert for spend caps) that says what happened, why it was a risk and what was done about it.
A threatening request is stopped before any model sees it
- A member types a requestScreening
"Write to the supplier and say we know where their staff live unless they drop the price by Friday."
Built for the buying committee that has to sign off.
The controls security, IT, and compliance expect, so adopting AI never means giving up control.
Mapped to the frameworks
One audit trail and evidence you can map to SOC 2, ISO/IEC 42001, the EU AI Act, and NIST AI RMF. Answer "how do you govern AI" with proof, not a promise.
Tamper-evident by design
Every enforcement record is hash-chained to its predecessor and verifiable offline. Any alteration breaks the chain and names the exact record.
Managed models, tokens at cost
Co-workers run on managed models. CortexHub runs inference and bills tokens straight through at cost, no markup, with a full per-model breakdown.
Isolated per organisation
Policy, credentials, audit, and usage are a hard boundary per organisation. One team's data is never reachable from another's context.
Runtime governance, plus a governed agent.
Tools that watch from above inventory and score your AI, but they do not sit in the path of a call. Automation platforms lead with workflow and add governance later. CortexHub governs at the point of action and gives you a governed Co-worker to deploy, under one policy and one audit trail.
Policy evaluated before every tool call, in the path of the action, never after the fact.
External agents and Co-workers answer to the same rules, vault, and audit trail.
Risky calls pause for human review, then proceed or are denied. Block-by-default is the fallback.
Secrets, and personal data your patterns match, are redacted before storage. Each team reaches only what you allow.
Records are hash-chained and verifiable offline, so evidence stands on its own.
An AI Co-worker with a Brain, on the same policy, not just a view of the AI you already have.
One audit trail and evidence you can map to SOC 2, ISO/IEC 42001, the EU AI Act, and NIST AI RMF.
Advanced compression shrinks large tool results in every prompt CortexHub builds, automatically, so you spend fewer tokens with no configuration.
Bring your own agents under governance.
The CortexHub Python SDK wraps the agents you build in a governed session. Point your framework at the session and every tool call is governed and audited, with no change to your agent logic.
- One governed session per run and end user, with the MCP URL and headers your framework already understands.
- Works with the OpenAI SDK, LangGraph, CrewAI, the Claude SDK, or any MCP client, no rewrite.
- Every tool call runs through the same policy, approvals, redaction, and audit as the rest of CortexHub.
pip install cortexhub# Python SDK - one governed session per run
from cortexhub import Cortexhub
cx = Cortexhub(api_key="cxh_key_...") # an API key bound to your agent
s = cx.session(agent="support-bot", mcp_session_id="conv-42",
end_user_subject="user_42")
# s.mcp.url + s.mcp.headers -> your MCP framework
# Or add to Claude Desktop / VS Code (GitHub Copilot):
# https://mcp.cortexhub.ai/v1/mcpEvery tool your teams use, governed the same way.
Curated toolkits ready to connect, plus your own business systems, governed from day one.
SlackGitHub
JiraNotion
Confluence
Google DriveHubSpotSalesforce
Linear
Airtable
Asana
ZendeskStripe
Shopify
Figma
Datadog
Intercom
PagerDuty
Snowflake
Zapier
Sentry
Dropbox
Any REST API you bringCommon questions
Start with one workflow
Put your first agent to work, under your rules.
Tell us what you want your agents to do and we will help you choose the right first workflow.
For the enterprise
See it govern your agents.
Book a demo for a walkthrough of runtime governance, the audit trail, and governed agent workflows. We will tailor the walkthrough to your tools, policies, and security requirements.