C3

Cloud Claude Code

An AI agent that runs on your dev machine and does your work while you sleep.

Not a cloud sandbox. Your always-on machine, your credentials, your real tools. Slack or Discord alerts trigger Claude Code (or Codex, or OpenClaude) sessions in your repos. The agent investigates, drafts a PR, and reports back. You review, merge, continue from your phone.

Your Sessions, From Anywhere

Every session is visible in one place. Agents that ran overnight, debugging sessions from your phone, all in one list.

Demo
C3
pulse-apiRunning2m ago
Investigate the spike in /api/metrics latency and optimize the aggregation query
c3Completed1h ago
Add PostHog analytics to the marketing layout with environment detection
nexus-appCompleted3h ago
Fix the webhook signature validation failing on large payloads over 1MB
orbit-docsError5h ago
Migrate search from Algolia to local vector embeddings with pgvector

Remote Sessions

Pilot Claude Code from your phone, tablet, or any browser. Your machine does the work. You just steer.

Autonomous Triggers

A Slack message starts an investigation. A Discord alert triggers a fix. Agents run while you sleep, and report back.

Real Conversation, Real Tools

C3 streams Claude Code's full output: thinking, tool calls, file edits, test results. Everything you'd see in the terminal.

Demo
← Sessionspulse-apibypassRunning
U
Investigate the spike in /api/metrics latency and optimize the aggregation query
▼5 ops
▶🔍 Grep "handleWebhook" in src/api/
▶📄 Read src/api/webhooks/handler.ts
▶📄 Read src/lib/crypto/verify.ts
▶✏️ Edit src/lib/crypto/verify.ts
▶$ npm run test -- --grep webhook
C

Found the issue. The aggregateMetrics() function runs a full table scan on every request because the created_at index was dropped in the last migration.

Two changes made:

  1. Restored the composite index on (created_at, metric_type)
  2. Added a 5 minute cache with stale-while-revalidate for the aggregation endpoint

Query time dropped from ~1200ms to ~8ms. All tests pass.

Done · 12 turns · $0.43
Send a follow-up prompt...

How It Works

C3 Architecture: Slack/Discord alert triggers C3, which auto-starts a Claude Code session that investigates, fixes, and reports back. You monitor from your browser.

Not a Sandbox

Anthropic runs agents in their cloud. C3 runs on yours. Both are useful. They solve different problems.

CapabilityAnthropic Routines / Managed AgentsC3
Where it runsAnthropic cloud sandboxYour always-on dev machine
Production databaseNo direct accessRead-only via your query scripts
Cloud CLI and secretsScoped connectorsAzure CLI, kubectl, Key Vault via managed identity
Session continuityRun-to-completionResume on web, phone, terminal, VS Code
Model choiceClaude onlyClaude Code, OpenAI Codex, OpenClaude, anything
Prompt and behaviorAnthropic-defined shapePlain markdown in ~/.c3/prompts/

Agent proposes. Human disposes.

On enterprise SaaS with real paying customers, the agent does not touch production unilaterally. It reads the alert, investigates in the correct repo with the correct credentials, writes a PR with root cause, fix, tests, and recommendation, and stops.

You review, merge, and ship.

The agent saves investigation time. Not judgment time.

Safety lives in the prompt, not the platform. You write the rules per channel.

Running in Production

Not a pitch. Running continuously on a real paying-customer SaaS.

20
Days live
17+
PRs drafted
0
Off-hours wakeups
100%
Human reviewed

Full minute-by-minute walkthrough: CTO meetup presentation.

Built with Real Components

Everything on this page is rendered using actual C3 components. Fork it and make it yours.

RunningIdleCompletedError12 turns$0.43Bypass Mode

Up and Running in 4 Steps

1
git clone https://github.com/Ideaplaces/c3 && cd c3 && npm install
2
cp .env.example .env.local    # add your API keys
3
npm run build && npx pm2 start ecosystem.config.cjs
4
open http://localhost:8347    # done

Open Source

Apache 2.0. Fork it. Customize it. The landing page comes with it.

Built by

Ciprian Rarau

at IdeaPlaces

GitHubBlogLinkedIn