Regular Chat
One model handles the task from prompt to result. The fastest path for clear, focused work.
Best for: quick fixes and well-scoped tasks
Multi-model coding workflows · your providers · your machine
Put the right model on
every part of the job
Use one model for focused work, ask several to investigate a hard problem, pair a builder with a reviewer, or let a stronger model plan and review while a lower-cost model implements.
7-day free trial · No credit card required · Plans from $9.99/mo
Refactor the authentication flow and add regression coverage for expired sessions.
Regular Chat
Round Table
Pair Programming
Mirror
Pair setupDelegation
OrchestratorPlanning & review
Strong reasoning model
ImplementerWriting code
Faster, lower-cost model
LIVE WORKFLOW
Plan
2ImplementIn progress
3Review
Works with the models you already pay for
Claude
Codex
Gemini
Kimi
MiniMax
DeepSeek
Z.ai
OpenRouter
OpenCode
Claude
Codex
Gemini
Kimi
MiniMax
DeepSeek
Z.ai
OpenRouter
OpenCode
Choose the working style
One task. Four ways to run it.
Start simple, then add independent investigation, continuous review, or direct comparison only when the task benefits from it.
Round Table
Several models investigate independently, then a lead model reconciles their findings into one result.
Best for: hard decisions and ambiguous bugs
Pair Programming
One model builds while a read-only partner plans or reviews. Delegation adds a lower-cost implementer preset.
Best for: higher-confidence implementation
Mirror
Send the same task to two models and compare their uncombined answers side by side.
Best for: comparing approaches and providers
Delegation
Spend premium tokens on judgment, not every keystroke.
Put a stronger model in charge of investigation, planning, and review. It dispatches implementation tasks to a faster or lower-cost model, checks the result, and sends work back when it falls short.
You choose both models. Savings vary by provider and task, but the goal is simple: use fewer premium-model tokens without removing premium-model oversight.
01
Plan
The orchestrator investigates and breaks the task into guarded implementation briefs.
02
Implement
The lower-cost model performs the dispatched coding work inside those boundaries.
03
Review
The orchestrator checks the diff and tests, then approves, fixes, or sends it back.
Round Table
Put your best models in one room
Hard problems deserve more than one opinion. Round Table appoints a judge that briefs several seats — each a different model — lets them investigate the same problem in parallel, then synthesizes everything they find into a single, sharper answer.
Work first, review after
Seats each take a pass at the task, then the judge reconciles their work into one result.
Review first, synthesize after
Seats independently analyze before anyone commits, then the judge writes the verdict.
- Mix any providers as seats
- Live phase status as it runs
- Seats shown in the sidebar tree
- Re-run a pass with /roundtable
Briefing the seats
JudgeDistributing the brief
Claude
QueuedGemini
QueuedKimi
QueuedCodex
Queued Pair Programming
Two models. One builds, one reviews.
Pairing makes engineers better — it works for agents too. A driver plans and writes the code. A read-only critic forms its own approach, challenges the plan, then reviews the real diff — looping until they agree. You watch both work, side by side.
Plan & Critique
Both models plan independently, align on an approach, then the driver builds it.
Critique Only
The driver runs; the critic reviews each step. Lighter weight, faster loops.
Delegate Implementation
A stronger orchestrator plans and reviews while a faster or lower-cost implementer writes the code.
- Mix providers — e.g. Claude + Gemini
- Optional human approval gate
- Critic is read-only — no edit conflicts
- Tune how many rounds they iterate
Driver
full agent · writes code
Critic
read-only · reviews
- Critic · read-only
Critique
The critic reviews the approach and requests changes.
- You
Your call
Optional human gate — approve, tweak, or send feedback.
- Driver
Build
The driver writes the code. It is the sole writer.
- Critic · read-only
Review the diff
The critic reviews the actual changes, read-only.
- Converged
Ship
They converge — you get reviewed, working code.
Everything in one place
Every provider, every platform, one interface. You own it, you run it.
Every Provider
- Claude, Codex, Gemini, and Kimi via their native CLIs
- MiniMax, DeepSeek, Z.ai, OpenRouter, and OpenCode-backed models
- Subscription-compatible: bring the plans you already pay for
- No model fees from us. You pay providers directly.
Every Platform
- Native desktop apps for Windows, macOS, and Linux
- Full iOS and Android mobile apps
- Control desktop agents from your phone in real time
- One account, every device
Session Control
- Run unlimited sessions across projects at once
- Resume any session right where it left off
- Move a live conversation into another provider with carried context
- Monitor context, token usage, cost, and reset windows
- Watch streaming output and interrupt instantly
- Switch models mid-session
Remote & Mobile
- One-button Cloudflare Tunnel — no port forwarding
- QR-code pairing for instant mobile connect
- Encrypted tunnel over any connection
- Send yourself host files as remote download cards
- Co-work from another remote client on the same live session
- Push notifications when agents finish or need you
Automation
- Schedule agents daily, weekly, or monthly
- Queue follow-ups for provider reset times
- Wake up to completed work
- Task history and output logs
- Automatic session resume on failure
Privacy & Security
- Fully self-hosted — the server runs on your machine
- Files never leave your computer
- Only the providers you choose ever see your code
- Agents ask before touching files or running commands
Voice dictation
Speak to your agents hands-free with Whisper, included free.
Reset-time follow-ups
When usage limits hit, schedule the next message for the reset window.
Permission control
Approve or deny file and shell access in real time.
Your agents, in your pocket
Agents run on your desktop. Watch them, steer them, and approve actions from your phone — anywhere, over a secure one-button tunnel.
Windows macOS Linux iOS Android
Up and running in minutes
01
Install the desktop host
Download for Windows, macOS, or Linux and choose Run agents on this computer.
02
Connect one provider
Sign in to Claude, Codex, Gemini, Kimi, or another provider you already use.
03
Run one focused task
Start with Regular Chat, then add Round Table, Pair, Delegation, or Mirror when useful.
One price. Everything included.
Every plan includes a free 7-day trial. No credit card required.
Monthly
$9.99/month
Full access, cancel anytime.
- Unlimited agent sessions
- All platforms: Windows, macOS, Linux, iOS, Android
- Supported providers: Claude, Kimi, Codex, Gemini, MiniMax, DeepSeek, Z.ai, OpenRouter, OpenCode
- Regular Chat, Round Table, Pair Programming, Mirror, and Delegation workflows
- Real-time session control
- Session persistence and resume
- Usage monitoring and reset-time scheduling
- Cross-provider context transfer
- Remote file handoff and co-working
- Scheduled tasks and automation
- Lifetime updates
- Priority support
7-day free trial · No credit card required
Most Popular
Annual
$49.99/year
Save $69.89/yrBest value. Save over 58%.
- Unlimited agent sessions
- All platforms: Windows, macOS, Linux, iOS, Android
- Supported providers: Claude, Kimi, Codex, Gemini, MiniMax, DeepSeek, Z.ai, OpenRouter, OpenCode
- Regular Chat, Round Table, Pair Programming, Mirror, and Delegation workflows
- Real-time session control
- Session persistence and resume
- Usage monitoring and reset-time scheduling
- Cross-provider context transfer
- Remote file handoff and co-working
- Scheduled tasks and automation
- Lifetime updates
- Priority support
- Early access to new features
7-day free trial · No credit card required
Frequently asked questions
Round Table sits several models around one problem. A judge model expands your request into a brief, hands it to multiple seats — each a different provider or model — and they investigate in parallel. The judge then synthesizes one answer from all of their work. You pick the seats; they can be Claude, Gemini, Kimi, Codex, and more, all at once.
Pair Programming pairs two models like two engineers. A driver does the planning and is the only one that writes code; a read-only critic forms its own opinion, reviews the plan, then reviews the actual diff — iterating with the driver until they converge. You can mix providers (e.g. Claude as driver, Gemini as critic) and add an optional human-approval gate before any code is written.
Delegation keeps a stronger model in charge of investigation, planning, and review while a faster or lower-cost model handles the implementation tasks it receives. This shifts routine implementation away from your most expensive model without removing senior-model oversight.
Mirror sends the same task to two models at the same time and shows their uncombined results side by side. Use it when you want to compare approaches or decide which provider is better suited to a task.
Yes. You log in to each CLI like you normally would. Agents Dispatch uses the native provider harness underneath, so your existing Claude, Codex, Gemini, or Kimi subscription just works.
Pretty easy. Run Agents Dispatch on your desktop, tap the one-button Cloudflare Tunnel installer on the Mobile Connect screen, then open the app on your phone and log in. That's it.
No. The server runs on your own machine. Your files never leave your computer. Mobile and remote access just lets you talk to agents running on that machine from wherever you are.
We don't provide models. You use your own. Claude, Codex, Gemini, and Kimi work with your existing subscription through their native CLIs. API-key and OpenCode-backed providers cover MiniMax, DeepSeek, Z.ai, OpenRouter, and more.
No credit card is required to start. When the 7-day trial ends, subscribe from the billing portal to keep access.
