AI Agent Orchestration

10× more tasks.
Same team.

Run 5, 10, 20 AI agents in parallel — each on its own task, its own branch, in an isolated container. You stay in control. Every decision is yours.

Request Early Access See how it works
US-142 coding
🟢 auto
US-187 review
🟡 checkpoint
US-203 coding
🟢 auto
US-211 done
merged
US-142 · feat: quiz validation layer US-187 · chore: schema migration ready for review US-203 · style: dark mode token update US-211 · merged to master ✓ US-142 · feat: quiz validation layer US-187 · chore: schema migration ready for review US-203 · style: dark mode token update US-211 · merged to master ✓
The Problem

Parallel agents without
orchestration is chaos.

You've tried running multiple Claude Code sessions. They collide — overwriting each other's changes, rebuilding shared code, stepping on each other in the same repo.

01
Agents collide
Two agents touch the same shared package simultaneously. One overwrites the other. Neither knows it happened.
02
No visibility
Five terminal windows open. You don't know which one is working, stuck, or silently broken until you check manually.
03
Merge anxiety
Agent finished — but you have no idea what it changed or how it affects the rest of the codebase. Every merge is a gamble.
Lifecycle

From issue to merged code
without touching a terminal.

Write a task in the UI, create an issue in Linear, or let a Sentry alert fire. 10times.dev handles everything else — cloning, branching, running the agent, rebasing, notifying you, previewing, merging.

📋
01
Create Task
Tasks tab, Linear issue, or Sentry alert
⚡
02
Auto Spawn
Clone + container + ports in ~30s
🤖
03
Agent Works
Reads TASK.md, codes, commits
🔔
04
You're Notified
Telegram + Linear status update
👁️
05
Preview
Local or remote public URL
?
06
Your Decision
Merge, reject, or give feedback
🔄
07
Auto Rebase
Branches stay current, conflicts auto-resolved
The Principle
You are of more value
than many agents.

AI doesn't know good from mediocre. It can't feel that the UI is off by just one pixel in a way that kills conversion. It doesn't have 10 years of context about your users, your team, and your product.

With 10 agents working in parallel, your judgment — what to build, what to reject, when it's good enough — becomes the bottleneck. Your taste is the compiler. Your experience is the architecture.

"You are of more value than many agents" — Mt 10:31
Platform

Everything you need to
orchestrate at scale.

🔒
Full Isolation
Every agent runs in its own Git clone with its own branch, file system, and zero access to your host environment. What happens in the agent stays there.
~/project/.10timesdev/agents/
├── PROJ-1/git/ ← own clone, own branch
├── PROJ-2/git/ ← own clone, own branch
└── PNL-39/git/ ← own clone, own branch
👁️
Live Preview — Local & Remote
Every branch, every agent — instant preview before the PR exists. Local with hot-reload, remote with public URL for clients and testers.
Mode What you get For who
Local Allocated ports · hot reload · see it while it's building You, fast review
Remote Netlify/Vercel deploy + Supabase branch DB · public URL · auto TTL 24h Client, tester, PM
📝
Tasks Tab
Write tasks in a markdown editor with image support — paste screenshots with Ctrl+V, drag & drop images, they're embedded directly in the task. Separate multiple tasks with ___. Agents pick them up, status shows inline. Done agents auto-merge to main.
📌
Multi-Tracker Integration
Not locked to one issue tracker. Pluggable architecture — use what you already have.
Local (SQLite)
Linear
Sentry
+ your own
🧠
AI Rules
Configurable rules injected into agent context. Global rules for all agents, project rules for specific repos. The agent reads them and decides which to apply based on task and codebase.
🔄
Auto-Rebase
Branches don't go stale. The monitor checks regularly — agent stopped and behind main? Rebase automatically. Conflict? Wake the agent with the conflict file list. The merge at the end is always clean.
🚨
Sentry → Auto Fix Pipeline
Production alert fires → agent spawns with full context: stacktrace with source code, breadcrumbs, URL, environment. It sees the exact file, line, function, and user actions that led to the crash. Fix lands in review — no manual triage.
💬
Wake on Comment
Two-way communication through your tracker. Write a comment — agent wakes up with your instructions. Agent posts "Ready for review" — you reply with feedback, it iterates. The tracker is the conversation channel.
🗨️
Agent Chat
Direct chat with running agents from the UI. Send instructions, paste screenshots, see responses in real time. Works alongside tracker-based communication.
📱
Telegram Notifications
Dedicated thread per task. Every commit, every status change, every "ready for review" — straight to your phone. No dashboards required.
🌐
Local & Remote Projects
Two modes, automatic. Local — no remote, agents clone and merge locally. Remote — full GitHub integration with PRs and branch management. Clone, push, merge, rebase all handle both modes transparently.
🏠
Self-Hosted
Runs on your machine. Your code never leaves your environment. No SaaS dependency, no API proxying, no data sent anywhere except to the AI provider and your own integrations.
Commit Protocol

Agents don't stop.
You stay informed.

Two-tier commit classification means agents never block on a review. They mark what needs your attention and keep moving. You catch up asynchronously.

🟢 AUTO
Refactoring, styling, new files, obvious bugfixes Agent continues without waiting. No review needed.
🟡 CHECKPOINT
API changes, schema migrations, shared packages, config Requires your review. Agent pushes, marks, moves on. You're notified.
🟢 AUTO
Component isolation, test additions, type fixes Safe to ship. Auto-classified.
🟡 CHECKPOINT
New migration file: add_quiz_results_table.sql Full description: WHAT + WHY + IMPACT + SQL attached. You decide when to run.
Database rule Agents NEVER run migrations. They create SQL files in supabase/migrations/ and mark 🟡. You run them manually on stage, then prod. Always.
Conflict rule Agents rebase on master every 3–5 subtasks. Push after every subtask. Never force-push without --force-with-lease. Never touch another agent's code.
Scale

Designed to grow with you.

100
concurrent agents
100 port slots × 6 ports each. Shared across all your projects.
0
extra infra
SQLite + JSON. No cloud dependency, no external service, no ops overhead.
24h
preview TTL
Remote previews auto-expire. No forgotten Netlify deploys eating your quota.
∞
projects
One instance. Multiple repos, tracker workspaces. Pluggable AI — Claude Code, Aider, or bring your own.
Conference Talk

"You are of more value
than many agents."

A talk about what nobody in the AI hype cycle wants to admit: more agents means more human judgment, not less. Your taste, context, and intuition become the system's only meaningful bottleneck.

Live demo. Real repo. Real agents running in parallel. No slides pretending this works — it works on stage.

50 MIN Full format — 6 blocks + live demo + Q&A
25 MIN Condensed — one demo, one thesis
WORKSHOP 4–8h — attendees build on their own repo
Target conferences
BoilingFrogs WarsawJS 4Developers Infoshare Confitura Segfault JSConf AI Engineer Summit LeadDev
Speaker

Former academic lecturer. Solo developer. Founder of UkryteSkarby.pl. Building in public with Claude Code since it shipped. This talk is live documentation of a real system running a real product.

Early Access

Run your first
10 agents today.

10times.dev is in private early access. Join the list and get the repo, the CLAUDE_GLOBAL.md, and a walkthrough of the full setup.

No pitch. You'll get a GitHub link and a setup guide.