Claros Cortex
One inbox for everything you need to do.
Cortex is a local-first desktop application that pulls tasks from Outlook, Teams, GitHub, Jira, Confluence, Slack, and IMAP into one prioritized dashboard. AI advisor personas draft context-aware responses. Nothing sends without your approval. All your data stays on your machine.
Your work is scattered across seven tools. None of them talk to each other.
You start the morning in Outlook. Twelve unread emails, three of which need responses. You switch to Slack. A thread from last night needs attention. GitHub has two PR reviews waiting. Jira has a ticket someone assigned overnight. Teams has a message from your manager.
Each tool is a separate context switch. The same issue appears in three of them — a GitHub PR mentioned in a Slack thread and linked in a Jira ticket — but you process it three times because nothing connects them.
By noon you've been reactive for four hours. The work that actually moves things forward hasn't started. This isn't a productivity problem. It's a tooling problem. Your work is fragmented and nothing aggregates it.
The approach
Three things that matter
Local-first
Every byte of data stays on your machine in SQLite. No cloud backend, no external database, no data leaving your device except through actions you explicitly approve. Your work context is yours.
Human-in-the-loop
Read operations run freely — polling email, fetching PR status, checking Jira. Every write operation requires your sign-off before it executes. This isn't a preference toggle. It's enforced at the system's dispatch layer.
AI advisors, not a chatbot
Five domain-specialized personas — CTO, CFO, PM, HR, Legal — each with scoped knowledge, distinct system prompts, and independent learning profiles. The CTO advisor sees architecture docs. The Legal advisor sees compliance docs. They learn your preferences separately.
Step by step
How Cortex works
Connect your tools
Seven integrations: Outlook (Graph API), Teams (Graph API), GitHub (REST + GraphQL), Jira (REST API), Confluence (REST API), Slack (Socket Mode), and IMAP for any other mailbox. Each supports multiple instances — connect work and personal accounts separately.
Automatic classification
Every incoming task passes through a 10-stage pipeline: allowlist check, recurring pattern detection, user filter rules, noise/bot detection, keyword classification, AI classification, advisor assignment, deadline extraction, suggestion trigger, and quarantine. Tasks are fully classified before they reach your dashboard.
AI advisor drafts a response
For important tasks, the assigned advisor assembles context from five layers: advisor personality, contact intelligence (learned communication style for this specific person), behavioral learning (your preferences), scoped knowledge documents, and per-advisor learned preferences. Then it drafts a response in the correct format — email reply, code review comment, Jira comment in ADF, Slack thread reply.
You approve, edit, or reject
Every AI suggestion waits for your action. Approve sends it through the originating integration. Edit lets you refine before sending. Reject teaches the system for next time. The approval rate feeds back into the advisor's learning profile.
The system learns
Behavioral learning tracks your priority preferences per source, preferred reply lengths, and active hours. Contact intelligence builds cross-channel profiles — the same person on email, Slack, and GitHub is linked into one identity. Per-advisor preferences improve over time.
Interactive demo
See it in action
An engineering manager's dashboard: tasks from GitHub, Slack, Jira, and Outlook — deduplicated and prioritized. The CTO Advisor drafts a code review comment. The PM Advisor drafts a stakeholder reply. One approval, one edit, one reject.
What Cortex can do
Capabilities
Seven Integrations
Outlook, Teams, GitHub, Jira, Confluence, Slack, and IMAP. Multi-instance support for each. 14 agent tools for GitHub alone (PR management, code review, CI status, issue CRUD). 50+ additional integrations on the roadmap.
Five AI Advisor Personas
CTO, CFO, PM, HR, and Legal advisors with distinct system prompts, knowledge scopes, and classification rules. Fully customizable — create new roles, edit prompts, adjust knowledge access.
Agent Chat
Conversational AI agent for multi-step task execution. Type natural language commands: "Find all high-priority GitHub issues assigned to me this week and draft a status update." Reads execute automatically. Writes pause for approval.
Action Palette
User-initiated write actions across any integration from any task card. Six intent categories (Track, Message, Document, Notify, Review, Status) with 25 tagged tools. Pre-filled forms from task context.
Contact Intelligence
Cross-channel identity resolution links the same person's email, Teams, GitHub, Slack, Jira, and Confluence accounts into one profile. Learns communication style per contact: formality, greetings, closings, preferred reply length.
Behavioral Learning
Tracks priority preferences per source, preferred reply lengths, active hours, and suggestion approval patterns. Feeds into context assembly so AI suggestions improve with every interaction.
Knowledge Index
Local repository of markdown documents, chunked and indexed with full-text search. Named entity extraction identifies people, projects, and systems. Each advisor role accesses only its scoped subset of documents.
Notification System
Rule-based alerts to Slack channels and SMS. Configurable priority thresholds and category filters. Timezone-aware scheduling — only notify during working hours.
Supported integrations
Connect your tools
| Integration | Protocol | Read | Write |
|---|---|---|---|
| Outlook | MS Graph API | Email polling, directory search | Replies, mark-read |
| Teams | MS Graph API | Chat polling, mention detection | Replies |
| GitHub | REST + GraphQL | Notifications, PR/issue status, CI | Comments, reviews, issue management |
| Jira | REST API v3 | JQL polling, issue details | Issue CRUD, transitions, ADF comments |
| Confluence | REST API v1/v2 | CQL polling, page content | Page CRUD, comments |
| Slack | Socket Mode | Channel ingestion, DM commands | Thread replies, Block Kit responses |
| IMAP | Standard IMAP | Any mailbox (Gmail, Yahoo, self-hosted) | — |
Remote access
Access from anywhere. Data stays on your machine.
Cloudflare Tunnel
Zero-config public HTTPS URL to your local Cortex instance. HMAC-SHA256 challenge-response authentication — the passphrase never traverses the wire.
Tauri v2 Desktop Client
Lightweight native app for remote devices. Scan a QR code on your Cortex server to pair. Server credentials stored in a Stronghold encrypted vault.
Slack DM Interface
Message the Cortex bot from any device with Slack. Outbound-only WebSocket — no open ports. Agent responses render in Block Kit with Approve/Edit/Cancel buttons.
Installation
Install and run
Windows Installer
Single-file installer with system tray app, auto-start on login, and optional Windows service mode. Download and run.
Docker Compose
Containerized deployment for Linux, macOS, and headless servers. Mount your data directory and configuration.
Tauri v2 Client
For remote access from a second device. Pairs with your running Cortex server via QR code.
PWA
Installable from the browser with offline support. For environments where a native install isn't possible.
System requirements: Python 3.12+, 4GB RAM minimum, SQLite. Runs on any machine that can run a local web server.
The platform
Part of the Claros platform
AI advisor suggestions are grounded in Scribe's knowledge index. Documentation referenced in suggestions links back to the source.
Process-related tasks in Cortex link to the relevant Cadence process for context.
Procedure execution reminders and runbook-related tasks surface as prioritized items. Execution deadlines appear in Cortex's timeline.
Handoff items with approaching deadlines surface as prioritized tasks. Flagged items generate notifications.
Decision review reminders and pending approvals appear as tasks. Upcoming evaluation deadlines are tracked.
Who uses Cortex
Built for
Engineering Managers
Unified view of PRs, tickets, threads, and emails across your entire team. AI drafts code review comments and stakeholder updates grounded in your architecture docs.
Technical Leads
Stop context-switching between six tools every morning. Deduplicate cross-channel noise. Focus on the work that actually moves things forward.
Consultants
Multiple client accounts across multiple tools. Multi-instance support means each client's Jira, Confluence, and Slack stays organized in one view.
Anyone drowning in notifications
Seven inboxes become one. AI handles the routine responses. You handle the ones that need a human.
Your seven inboxes. One dashboard.
Download Cortex and run it on your own machine. Your data never leaves your device.