For macOS Sequoia (15) & later

Many minds.
One reply.

Polycode fans every prompt out to Claude, GPT, Gemini, Apple Intelligence and more — in parallel — then synthesizes a single answer with the trace kept honest. Native macOS. Bring your own keys.

Notify me at launch
Lock in launch pricingComing soon to the Mac App StoreUniversal · Apple Silicon + Intel
New session— polycode-apple
polycode-apple
Search⌘F
polycode-apple
New session
— in flight
Flaky test triage
Consensus · 6
SwiftData migration
Opus 4 · 8
consensus-engine
Provider golden fixtures
Gemini · 5
MCP registry filter UX
Consensus · 4
MCP · 3 connected
What's the best way to handle a flaky test in CI?
Consensus of 4
A
Claude Sonnet 4.6
O
GPT-5
G
Gemini 2.5 Pro
Apple Intelligence
Quarantine first, then triage.
A
Move the failing test to a flaky lane that still runs but doesn't gate the merge — that stops the bleeding without hiding the signal.
A
O
Then write a deterministic repro: pin time, seed randomness, and isolate the network or filesystem dependency it’s racing on.
A
O
Sources
A
Claude Sonnet 4.61240ms
O
GPT-5980ms
Apple Intelligence410ms
CopyRegenerateView trace2541 in · 1213 out · $0.0055
Reply to Polycode…
A
O
G
4 models·primary
A
Claude
#No mode
196 tok · $0.001Send
focuscancelfilestargetScoped to polycode-apple
Provider Trace
Consensus reached
3/4 providers · 1420ms · 94% agreement
A
Claude
O
GPT-5
G
Gemini
On-device
A
Claude Sonnet 4.6
claude-sonnet-4-6
1240ms
First tok
280ms
Total
1240ms
In / Out
847 / 412
Cost
$0.0024
O
GPT-5
gpt-5
980ms
First tok
210ms
Total
980ms
In / Out
847 / 489
Cost
$0.0031
Full response
Try retry(3, backoff: .exp) first — most CI flakes are transient network races.
G
Gemini 2.5 Pro
gemini-2.5-pro
timeout
— · no reply before the fan-out budget
Apple Intelligence
foundation-model-v1
410ms
First tok
95ms
Total
410ms
In / Out
847 / 312
Cost
Free

Polycode is in active development — this preview may not reflect the final state of the app exactly.

Providers

Works with the keys you already have.

A
Claude
O
GPT-5
G
Gemini
On-device
Ol
Ollama
X
xAI
OR
OpenRouter
Anything OpenAI-compatible
Enterprise
B
Amazon Bedrock
Az
Azure OpenAI
V
Google Vertex
OR
Or bring a single OpenRouter key — one signup covers most of the cloud.
11providers shipped
≤ 1.4stypical fan-out
0additional subscriptions
100%your keys, your billing
Plurality, not picking

Every prompt asks every model. Then a primary settles it.

One question fans out to as many providers as you've configured. Replies arrive in parallel — you watch them land — and a primary peer synthesizes them into a single answer. No more tab-juggling between Claude and GPT.

Fan-out · in flight
A
Claude Sonnet 4.6
✓ 1240ms
O
GPT-5
✓ 980ms
G
Gemini 2.5 Pro
timeout
Apple Intelligence
✓ 410ms
Synthesize → primary
tool-less · resolves disagreement
3/4 · 1.42s

Parallel by default

Reads (synthesis) fan out to every configured provider. Native writes are primary-only — your write tools never run twice.

A primary peer chooses the words

Pick the model whose voice you trust most. It reads every reply, then writes the consensus you actually see.

Disagreement is surfaced, not buried

When the panel splits, the agreement score says so — and every provider's full reply sits one click away in the trace, minority takes included.

Trace, don't hide

See who said what. Down to the token.

Hit ⌘⇧I and the inspector slides in. Each provider gets a card: status, latency, in/out tokens, dollar cost, the resolved model id, and the full response — ready to copy, share, or pin to disk.

Provider Trace· flaky test triage · turn 3
I
Consensus reached
3/4 providers · 1420ms · 94% agreement
A
Claude Sonnet 4.6
O
GPT-5
G
Gemini 2.5 Pro
Apple Intelligence
A
Claude Sonnet 4.6
claude-sonnet-4-6
✓ 1240ms
First tok
280ms
Total
1240ms
In / Out
847 / 412
Cost
$0.0024
O
GPT-5
gpt-5
✓ 980ms
First tok
210ms
Total
980ms
In / Out
847 / 489
Cost
$0.0031
Full response
Don't quarantine yet — first add a retry(3, backoff: .exp) wrapper. Most CI flakes are transient network races; if it survives 3 attempts, then isolate.
G
Gemini 2.5 Pro
gemini-2.5-pro
timeout
— · no reply before the fan-out budget
Apple Intelligence
foundation-model-v1
✓ 410ms
First tok
95ms
Total
410ms
In / Out
847 / 312
Cost
Free

Per-exchange, not per-session

Every fan-out renders its own trace card stack. Scrub back through your conversation; the receipts go with it.

Costs you can audit

Token-priced per provider with tabular figures. No mystery bills, no marked-up resale — just what your keys actually spent.

Tool calls included

Approve or reject individual tool calls per exchange, and the trace records which provider's plan was actually executed.

Built for the platform

A native macOS app — not a web wrapper.

SwiftUI from titlebar to inspector. Hardened runtime, App Sandbox, MAS distribution. Quick Look, drag-and-drop, system services, Spotlight indexing for sessions — it acts like the rest of your computer because it's built like the rest of your computer.

SwiftUI, top to bottom

26 / 22 / 17 / 15 / 13 — HIG type scale. Sentence case. Visual-effect titlebar. No Electron, no Tauri, no bloated web-shell chrome.

Sandboxed by default

App Sandbox + hardened runtime. Files are entitlements, not implicit reads. Polycode can't see anything you didn't grant.

On-device, when you want it

Apple Intelligence Foundation Models route everything through the Neural Engine — zero network, zero spend, sub-half-second turnarounds.

Spotlight-fast sessions

Sessions are indexed by Core Spotlight. ⌘F in the sidebar or just hit Spotlight from anywhere — every conversation is searchable.

MCP, first-class

Model Context Protocol servers attach as tools. Polycode shows status, capabilities, and per-server latency right in the sidebar strip.

Local-first, by design

Conversations live in a local SQLite store you can read, back up, or move. No cloud sync we control — your conversations stay on your Mac.

Bring your own keys

Your money. Your keys. Your prompts.

Polycode never sees your traffic. Keys live in the macOS Keychain, requests go directly from your machine to the providers you've authorized, and we couldn't read your conversations even if we wanted to — there's no server in the loop.

  • Keys stored in the system Keychain — never on disk in plaintext, never on our servers.
  • Direct calls to provider APIs. No proxy, no rewriting, no middleman markup.
  • Anonymous diagnostics are on by default and fully documented — toggle them off in Settings, no account required.
  • Run ollama, LM Studio, or any local OpenAI-compatible endpoint entirely offline if you want zero network at all.
Providers· 4 active · keychain
⌘,
A
Claude
sk-ant-•••••••••••••3a9f
✓ valid
validated · 4 models
O
GPT-5
sk-•••••••••••••••••2c0e
✓ valid
validated · 12 models
G
Gemini
AIza••••••••••••••••Vp1q
✓ valid
validated · 6 models
Apple Intelligence
on-device · Neural Engine
✓ valid
no key required
Ollama
http://localhost:11434
◇ local
reachable · 3 models loaded
OpenAI-compatible
add an endpoint…
+ add
optional
Launch price

Buy it once. Keep it forever.

No subscription. No per-seat. No usage tax. Pay once, then bring your own keys — you'll spend on the providers you actually use.

Polycode
macOS Sequoia (15) & later
$39one time
  • Universal binary for Apple Silicon and Intel Macs.
  • Unlimited providers, unlimited sessions, unlimited keys.
  • All future updates included. Forever.
  • Family Sharing supported. Up to 5 family members on one purchase.
  • Provider costs are billed by the providers themselves.
Join the waitlist for launch pricing
Coming soon to the Mac App Store.
Questions? Email about@polycode.me.
FAQ

Plain answers.

Why pay for a chat client when ChatGPT and Claude have native apps?
Because no single provider's app shows you what the other providers would have said. Polycode's whole shape is the comparison — fan-out, dissent, consensus. The official apps are great if you've already picked a horse; Polycode is for the times you haven't.
Do I need accounts with every provider?
No. Configure as many or as few as you want. With zero keys, you can still run Apple Intelligence on-device and Ollama against a local model. Most users start with one paid provider plus on-device, then add others as needed.
Can I just use OpenRouter for everything?
Yes, and for most people that's the fastest way to get the full fan-out experience. One OpenRouter key covers Anthropic, OpenAI, Gemini, xAI, Mistral, Meta, DeepSeek, and dozens more — one signup, one invoice, no juggling. The tradeoffs: OpenRouter sits in the network path between you and the upstream provider (so it's not as direct as a per-provider key), and most upstreams carry a small markup on top of the underlying API price. Apple Intelligence and local providers like Ollama or LM Studio aren't routed through OpenRouter — those still run on-device for free.
Will my prompts ever leave my machine through your servers?
No. Polycode talks directly to the providers you authorize. We don't run a proxy, we don't see your traffic, and we don't have a way to inspect, log, or train on your conversations.
How does fan-out cost compare to using one model?
It costs roughly N× one model where N is the number of configured providers — the same prompt is sent to each. The trace card shows exact tokens and dollars per provider, so you always know what you spent.
What about MCP servers and tool calls?
Polycode is a first-class MCP host. Connect a server, its tools appear inline on the consensus model, you can approve or reject calls per-exchange, and the trace records which provider's plan was actually executed.
When will an iPad / iPhone app ship?
Not yet on the public roadmap. The current app leans hard on macOS-only affordances — three-column layout, Inspector, Spotlight indexing, MCP processes — and we'd rather ship one native app well than three thin ones.

Longer answers live in the docs FAQ.

Stay up to date

Stay up to date with Polycode.

One email per release. No drip campaign, no growth hacks — just changelogs, new-provider announcements, and the occasional behind-the-scenes note.

  • First in line for the public launch on the Mac App Store
  • Eligible for early TestFlight invites as seats open up
  • No spam, unsubscribe in one click

Your address goes to our email service (Resend) and is used only for release notes — details in the privacy policy.

Talk to all of them.
Read just one.

Polycode is coming to the Mac App Store. Get on the waitlist to lock in launch pricing — or join the TestFlight beta to try it first.

Notify me at launch
Universal · Apple Silicon + IntelmacOS Sequoia (15) & later