Open source · runs in your browser

One prompt.
Every AI.

Relay sends your task to Claude, GPT, Gemini, AWS Bedrock and hundreds more at once. They check each other's work, and a judge merges the best answer.

Open Relay Try the demo, no keys needed
  • 5AI providers
  • 7orchestrations
  • 3cloud coding agents
  • 0servers

Press K anywhere to run the orchestra →

0agent calls
0tokens
0.0selapsed

    This is a simulation, and no AI is called here. Click any agent to see what it's told.

    • Claude
    • GPT
    • Gemini
    • AWS Bedrock
    • OpenRouter
    • Groq
    • Mistral
    • DeepSeek
    • xAI Grok
    • Together
    • KakaoTalk
    • Claude Code
    • Codex
    • Gemini CLI

    Three steps. That's it.

    1. 1

      Connect your AIs

      Paste the keys you have: Claude, GPT, Gemini, AWS Bedrock, or one OpenRouter key for hundreds of models. If a pipeline wants one you don't have, a connected AI stands in.

    2. 2

      Describe the task

      Paste any code or text the agents need. Start with ultracode: to throw everything at it, or say "ask all the models" for the Council.

    3. 3

      Press ⚡ Kickstart

      Relay picks the cheapest orchestration that fits and runs it. Every agent streams live, with tokens and cost, and the result can go straight to your KakaoTalk.

    Watch the router think

    ⚡ Kickstart reads your task and climbs only as high up the ladder as it needs to. This runs the same code as the app, so type anything.

      Seven ways to orchestrate

      Hover a card to tilt it, then press ▶ to watch that pipeline run in the orchestra above. In the app, every pipeline is editable, stage by stage.

      What will it cost?

      Drag the sliders. Higher rungs buy more checking with more calls, so start low and climb only when a task needs it.

      Rough estimate. Top means Claude Opus 5 ($5/$25) and GPT‑6 Astra ($10/$50) per million input/output tokens. Fast means Sonnet 5 ($2/$10) and GPT‑5.6 Terra ($2/$12). Reasoning tokens can add more.

      Cloud agents

      Real coding agents, side by side

      One click sends your task to GitHub Actions, where these agents work on it at the same time. Then Claude merges their answers.

      • Claude Codewith ultracode on: xhigh effort plus multi-agent workflows, through the Anthropic API or AWS Bedrock
      • OpenAI Codexat xhigh reasoning effort
      • Google Gemini CLIon the same task, in parallel

      Only the repository owner can start a run, and each agent runs only if its key is saved as a repository secret.

      Open Cloud run: design a production rate limiter #42
      1. Press play to watch a cloud run.
      agent-run
      ‹나와의 채팅
      Today
      ultracode: design a rate limiter 🙏

      Results in your KakaoTalk

      Connect KakaoTalk once and Relay messages you whenever a run finishes. Every final answer also gets a one-tap KakaoTalk button. Try the phone.

      🔒 No serverYour keys stay in your browser and go straight to each AI vendor.
      🧹 Keys stay localBy default they last only for the tab. Saving them is opt-in.
      🛡️ Locked downA strict Content-Security-Policy blocks other scripts, and every AI answer is sanitized before it's shown.
      💸 Spend guardsCap calls, fan-out size and concurrency, and stop any run instantly.

      Setup

      Each part takes a few minutes, and you only need the parts you want.

      🔑 Connect AI keys (in the app)
      1. Open the app, then click Settings.
      2. Paste any keys you have, then click Test & load models for each:
      3. Click Save, untick Demo, and press ⚡ Kickstart.
      ☁️ Turn on cloud agents (GitHub)
      1. Open the repo's Actions secrets page.
      2. Add any of these secrets. Each one switches on its agent:
        ANTHROPIC_API_KEYClaude Code (ultracode) + the merge step
        AWS_BEARER_TOKEN_BEDROCKClaude Code through AWS Bedrock
        OPENAI_API_KEYOpenAI Codex
        GEMINI_API_KEYGoogle Gemini CLI
      3. In the app, write a task and click Send task to cloud agents.
      💬 Connect KakaoTalk
      1. Create an app at developers.kakao.com.
      2. Under Platform → Web, add the site domain https://speddiikga-code.github.io.
      3. Under Kakao Login, turn it on and add the Redirect URI https://speddiikga-code.github.io/relay/app/.
      4. Under Consent items, enable "Send message in KakaoTalk" (talk_message).
      5. In the app, go to Settings → KakaoTalk, paste the REST API key, press Connect KakaoTalk, then Send test message.

      Questions

      Is this really Claude Code's ultracode?

      In the browser, Relay runs its own orchestration (the pipelines above) directly against the model APIs. In the cloud agents, it runs the real Claude Code CLI with the documented ultracode setting turned on. The typed ultracode: keyword works only when a person types it into Claude Code, by design, so no website can trigger it.

      What does it cost?

      Relay itself is free. You pay each AI provider for the tokens your runs use, on your own keys. The app shows tokens and an estimated cost per agent and per run. Try the calculator above.

      Can it read my files or my repository?

      The browser app can't. Paste in what the agents need. The cloud agents can read the repository they run in (read-only), and they answer in comments rather than pushing code.

      Can other people use my keys?

      No. Your keys are stored only in your own browser, so visitors to the site use their own. Cloud runs start only for the repository owner.

      Stop asking one AI.

      Put all of them on your task.

      Open Relay