Build an agent that belongs where people already work, talk, and live.
Overview · Get started · Templates · Coding agent · Resources
Build for Agents, Everywhere: Bots, Channels, & More, the AI Tinkerers global hackathon on September 12–13, 2026. Choose your city on the event page for its local schedule. Put an agent inside a conversation, an app, a phone, or a physical environment. Make the context of that place essential to what it can do.
This kit gives you three runnable templates, files to hand to your coding agent, and sponsor setup notes. Pick a user, a problem, and one complete interaction. You can use any stack; you do not need every sponsor or every surface.
Your project and its core functionality must be created during the event. Existing libraries, templates, and starter code are allowed; describe what you reuse and what you build. Read the rules, then follow your city's participant portal for the current deadline and judging criteria.
For web, paste this into your coding agent:
Help me get started with CopilotKit. Run this command and follow the instructions:
npx --yes copilotkit@latest onboard start
For Slack, follow Channels setup. For React Native, follow the Expo setup instructions.
These starting points serve different kinds of context. CopilotKit Channels brings the Slack agent into the conversation; CopilotKit React connects the web agent to the app people are using; CopilotKit React Native brings the same agent pattern onto a phone.
OpenAI + CopilotKit Channels + Exa
An agent reads what people already said, researches with Exa, and answers in the same thread with native cards and source links. Start with a support conversation, a research discussion, or a team decision.
The included Slack app supplies thread history, subscriptions, search, and Channels UI. Configure your model, Exa, and a managed Channel, then run npm run dev:slack. No public tunnel is needed. Teams or other chat platforms can use the same Channels pattern, but this starter ships the Slack app.
OpenAI + CopilotKit React + Ambiguous AI
An agent sees the page you are on and turns a request into a real workplace record you can still find after a refresh. Adapt it to customer follow-ups, a project workspace, or a personal planning app.
The included web app supplies page context, frontend tools, agent-rendered UI, and a browser approval step. Connect an Ambiguous AI workspace, then run npm run dev:web; approved follow-ups are saved through the server and can be read back after refresh.
OpenAI or OpenRouter + CopilotKit React Native
A mobile agent reads app state, renders native cards, and waits for a tap before changing local sample data. Start with a personal finance assistant, a field checklist, an inventory counter, or any workflow where phone context and approval matter.
The included Expo app supplies seeded finance state, native rendered tool UI, a human-in-the-loop expense approval, and a mobile-specific CopilotKit runtime endpoint served by the web app. Configure your model provider, start npm run dev:web, then run the mobile app from apps/mobile.
Use the React Native template →
The supplied on-call and finance assistants are infrastructure examples: read ambient context, call a tool, render useful UI, and return a verifiable result. Choose a different user, problem, dataset, and interaction; the goal is your own project, not another version of the starter scenario.
Use the demo prompts to learn how the pieces connect, then replace the sample domain. In the Slack sample incident flow, approval cards record decisions without executing production actions. In the web follow-up flow, the page approval button saves the reviewed Ambiguous task. In the mobile finance flow, approval changes local in-memory sample data. Enforce the same kind of write boundary around any external action you add.
Want another surface pattern? The web app also includes a voice route, and the shared agent can connect to remote MCP tools when configured. The event surfaces are inspiration, not separate tracks or a requirement to build multiple apps.
Give your agent these files before it starts coding:
| File | What it provides |
|---|---|
| hackathon-overview.md | The challenge, four surfaces, and official judging criteria |
| hackathon-rules.md | Build eligibility, inherited code, and required deliverables |
| using-sponsor-tools.md | Every sponsor featured in this kit: access, authentication, configuration, and a first working call |
| AGENTS.md | Repository conventions and verification commands |
| Channels skill | Verified Channels APIs for the Slack template |
The app READMEs provide launch commands, files to customize, and a concrete result to check. Start with one template and add a second surface only if it helps your user.
| Need | Go here |
|---|---|
| Event details, deadline, and judging | Find your city, then open its participant portal and handbook |
| OpenAI agent development | Agents SDK quickstart |
| OpenRouter access and model choice | Quickstart · Keys · Model catalog · Model switching |
| CopilotKit app development | Docs · Tools and context · Discord channel for technical questions |
| CopilotKit Channels | Channels guide · Screenshot walkthrough · OpenTag example app |
| Exa quickstart | Search API guide · Kit setup |
| Auth0 API authorization | Node API · Kit setup |
| Ambiguous AI quickstart | Developer guide · Kit setup |
| Rehearse and debug | Demo prompts · Troubleshooting |
| Prepare your entry | Submission checklist |
For credit redemption instructions, choose your city on the global event page and check its participant portal's Credits & Offers section.
For technical questions during the event, check your city's participant portal and ask your local organizers.
For the Slack/web workspaces, npm run verify runs typechecks and offline tests without credentials. The mobile app has its own install, tests, typecheck, and Metro export checks under apps/mobile. Each app reports missing configuration when the relevant integration is used. Live sponsor calls and platform delivery require your accounts. See developer docs for detailed setup and deployment.