supervise agents. don't replace yourself.

Jack in from
anywhere.

Your agents run on your machines. One dashboard for all of them, in real time. Free for everyone.

Get Started
my-projectClaude Code~/projects/my-project
refactor auth
fix tests
docs update
Add better error logging to the auth middleware
5Tool Callsall completed
Read
src/server/auth.ts
Grep
validateToken|AuthError
Edit
src/server/auth.ts
Edit
src/server/session.ts
Bash
npm test -- --filter auth
Done. I've added structured error logging to validateToken. Check the diff panel for changes.
shell
npm test -- --filter auth
output
exit 0
Tests: 14 passed, 14 total
Time:  1.2s
1.2s·4 lines
All 14 tests pass. Want me to commit
Ask anything or prefix with ! for shell...
Files
Git 3
Jobs
System
my-project·main
2
Changes3
Msrc/server/auth.ts
+
Msrc/server/session.ts
+
Asrc/types/auth.d.ts
+
History48
feat: add auth error logging and structured...
a3f2c1d2m ago
+6-2
fix: session timeout not resetting on activity
9b1e4a71h ago
+12-4
chore: update dependencies to latest
c7d3f093h ago
+88-34
feat: add token refresh endpoint
e2a8b1c5h ago
+42-8
fix: race condition in middleware chain
f4c9d2e1d ago
+18-6

zoom into the details

cortices.io
10%
Code quality watchdog
monitoring
ACTIVE
100
JS to TypeScript Migration
completed
DONE
100
Database Schema Audit
completed
DONE
35%
API v2 Rollout
analysis
ACTIVE
Right panel · Tasks tab

Long-Running Plans

Goals that span days, not minutes.

Define a multi-step goal. The agent breaks it into stages, tracks progress across cron-triggered runs, and picks up where it left off — with full conversation history.

  • Structured state across runs
  • Named stages with progress
  • Pause, resume, trigger manually
cortices.io
shell
npm run test -- --filter auth
output
exit 0
 PASS  src/auth/validate.test.ts
 PASS  src/auth/session.test.ts

Tests: 11 passed, 11 total
Time:  0.84s
0.84s·6 lines
shell$ npm run test
All 11 tests pass — go ahead and commit with “feat: add auth logging”
Chat · ! prefix

Shell Mode

Run commands right from the chat.

Prefix any message with ! to execute it on your agent's machine. Output streams live. The agent sees the result and can act on it.

  • Tests, builds, git — anything
  • Real-time streaming output
  • Kill long-running commands
Review·5 files1
src/server
src/server/auth.ts
oldnew
@@ -42,8 +42,12 @@
43const user = await db.find()
43const user = await db.find()
44if (!user) return null
44if (!user) {
 
45 logger.warn('Not found')
 
46 return null
 
47}
45if (user.suspended) {
48if (user.suspended) {
46 throw new AuthError()
49 throw AuthError({ userId })
Comment on line 49
Add suspension timestamp too
47}
50}
click line numbers to comment · drag to select rangeSubmit 1 comment
Right panel · Git tab

Review-Driven Workflow

Review code like a PR. But faster.

Every change your agent makes appears as a syntax-highlighted diff. Select lines, leave inline comments, and your agent iterates from your feedback on the next turn.

  • Full git diff with +/- highlighting
  • Inline comments on any line
  • Agent reads and addresses your feedback
cortices.io
API Health Checkevery 5 minCMD
Daily Git Summaryevery dayAI
Disk usage monitorevery 10 minAI
Log Rotationevery 6 hoursCMD
Stale Branch Cleanupevery weekAI
Right panel · Tasks tab

Scheduled Tasks

Automate the boring stuff.

Run recurring tasks as bash commands (free) or AI-driven prompts (Claude reasons through it). Visual run history with status dots.

  • Command mode: zero LLM cost
  • AI mode: full reasoning
  • Toggle on/off per task
cortices.io
preference
No semicolons in TypeScript
User prefers no semicolons. Applies to all TS/JS files.
scope: global · pinned
fact
Senior backend engineer, new to React
10+ years Go experience. Frame frontend explanations using backend analogues.
scope: global
learning
Shared cache between service-a and service-b
Clearing either side's cache affects both services. Be careful with cache invalidation.
scope: api-server
log
Auth module migrated from Express to Hono
2026-03-19: 5 files changed. Legacy Express middleware removed.
scope: my-project
Memory page

Cross-Agent Memory

Tell one agent. All of them remember.

Preferences, facts, learnings, and logs — stored once, shared across every agent and project. Scoped or global, pinned or contextual.

  • Four memory types: fact, preference, learning, log
  • Scope to specific projects or share globally
  • Pin critical context so it's always present

Mobile

Your agents in your pocket.

Open the dashboard in your phone's browser, tap the Share button, then “Add to Home Screen” — it installs as a native app. Run shell commands, review diffs, and chat with your agents from anywhere.

  • Progressive Web App — works offline
  • Full chat, files, and shell access
  • Add to Home Screen — launches like a native app
my-projectClaude CodeS1
Run the LoRA fine-tune on GPT-2 with the new dataset. Use lr=2e-4, 3 epochs.
4 tool callsEdit, Bash ×3
Training started on GPU 0. ETA ~18 min for 3 epochs on 12k samples. Loss after epoch 1: 0.847
🌊
42%vibing
Check GPU memory usage
1 tool callBash
GPU 0: 6.2 GB / 24 GB used (RTX 4090). Plenty of headroom — batch size could go up to 32.
what shall we build today?
✓ autoexec
Chat
Files
More

faq

Frequently asked questions

Cortices currently supports Claude Code (Anthropic) and OpenAI Codex. You can connect both simultaneously and manage them from one dashboard. OpenCode support is on the roadmap — reach out if you'd like to request a specific agent.

No. Agents connect outbound to the Cortices hub, so they work behind NATs, firewalls, and corporate networks — no port forwarding or VPN required.

Cortices relays messages between your agent and the dashboard. API keys and credentials are passed through directly to your agent machine — Cortices never stores them. Your code stays on the machine where the agent runs. File contents are only transmitted on-demand when you explicitly open a file in the viewer.

Cross-agent memory lets your agents share context — facts, preferences, and learnings — across different projects and machines. It helps agents stay consistent without you repeating yourself.

Scheduled tasks are simple cron jobs that run independently each time (like health checks or log cleanup). Plans are stateful, multi-step goals that track progress across runs — ideal for migrations, research, or phased rollouts.

Yes. Cortices is a PWA — add it to your home screen on iOS or Android and it works like a native app. Review diffs, chat with agents, and approve permissions from your phone or iPad.

Yes. Cortices is completely free and open to everyone. Sign up and start connecting your agents immediately — no invite code, no waitlist, no usage limits.

Yes. Cortices provides the interface to monitor and control your agents, but you are responsible for the commands and permissions you grant them. We strongly recommend reviewing agent actions before approving, and Cortices is not liable for any unintended changes made by your agents.

You set the direction. Agents do the work.

Free for everyone. No limits. Get started in 2 minutes.

Get Started