Free Build
Build an agent around a problem you care about. The guided support agent remains available if you want more structure, but this track gives you room to choose the job, instructions, integration, and demo.
Goal: create a working agent of your own that completes at least one real call through an external connection or integration.
Need some inspiration?
Explore official eve templates and examples from community builders for inspiration. Remember to choose a problem that interests you. You are not limited to the examples below. Keep your workshop version focused on one useful behavior and one connection.
| Inspiration | What it demonstrates | A workshop-sized version |
|---|---|---|
| Social media agent | An official Slack-based agent that drafts social content, pulls briefs from Notion, and publishes through Typefully | Turn one brief into a social post and send it to a publishing queue |
| Marketing team | An official multi-agent team that delegates work to specialists and publishes through external services | Use one specialist and one publishing integration to create a campaign asset |
| Agentic CRM (launch post) | A community CRM where an eve agent researches records, tracks evidence, and schedules follow-ups | Research one company or contact, save the findings, and flag uncertain details for review |
| AnyPay | A community payments agent that lets agents pay in USDC while merchants receive local stablecoins | Quote a payment or exchange outcome, then put the action behind human approval |
Capabilities to explore
The workshop does not require these features, but they can help you take a Free Build beyond one prompt and one integration. Add one only after the core agent loop works.
| Capability | What it does |
|---|---|
| Skills | Load-on-demand playbooks for repeatable procedures, such as a research method, writing style, or review checklist. Adds focused instructions without adding another tool or changing the agent's identity. |
| Subagents | Delegate work to a specialist with its own prompt and tool surface. Use one when part of the job needs real specialization or can run independently, such as a researcher gathering evidence before the main agent writes a report. |
| Schedules | Start the agent on a recurring cadence. Fits daily digests, weekly reports, data syncs, and other work that should run without waiting for a new message. |
| Sandboxes | Give the agent an isolated workspace where it can run commands, execute scripts, and read or write files. Every eve agent includes one by default, and you can customize it with setup, seed files, or tighter network access. |
eve now searches skills.sh through its built-in @skills source. Search for a relevant skill or add one you already know:
npx eve registry search react
npx eve add @skills/vercel-labs/agent-skills/vercel-react-best-practicesGo further with integrations
Any MCP connection works for Free Build—you are not limited to the examples below. Choose one that lets your agent retrieve real data or take a useful action for your demo.
Choose a channel where your audience already spends time:
Team and community chat
Slack, Discord, Microsoft Teams, or Google Chat.
Then give your agent data and actions beyond the model:
Database
Supabase gives an agent access to structured application data and database operations. Start with a read-only workflow, then place any writes behind human approval.
Memory
Mem0 can store and retrieve information across sessions, such as preferences, prior decisions, or research history.
Browser use
Browser Use, Browserbase, and agent-browser can research websites and interact with web pages.
Research and data
Similarweb, Notion, or Airtable can provide market intelligence, knowledge, or structured records.
External actions
Resend, Stripe, Razorpay, or Brex can let an agent send communications or work with financial systems.
Browse the eve integrations directory for more options.
Minimum outcome
A channel changes where people talk to the agent. For this challenge, also add a capability that lets the agent retrieve data or perform useful work outside the model.
Recommended build strategy
Define the job before the features.
Complete this sentence:
PromptMy agent helps [person] accomplish [job] by using [external system].Keep the first version narrow enough to demonstrate with one or two prompts.
Create a fresh agent or reuse your workshop project.
npx eve@latest init <your-agent-name>Replace
<your-agent-name>with whatever you want to call your agent. Reusing the guided project is also fine. Replace its instructions and remove capabilities that do not belong in your new use case.Find an integration.
npx eve registry list npx eve registry search githubSearch for the system your idea needs, then inspect the candidate before installing it:
npx eve registry view connection/<name> npx eve add connection/<name>You can also browse the eve integrations directory. Registry items may add dependencies and project files, so review the generated changes before running them.
Run locally.
npx eve devSend a prompt that needs your chosen capability and watch for the external tool call. If you added a skill, subagent, or schedule,
npx eve inforeports whether eve discovered it. If you customized the sandbox, test a prompt that uses its shell or file access.Test three behaviors.
Try one happy-path prompt, one prompt with missing context, and one request that should be refused or paused for approval. Save the strongest prompt for your demo.
Demo your agent
| Part | What to show |
|---|---|
| Problem | “My agent helps…” |
| Prompt | Send the prepared request |
| Integration | Point out connection_search or the external tool call |
| Result | Show the useful answer or completed action |
| Boundary | Show one safety choice, approval, or graceful failure |
Keep the demo focused on one complete loop. A small agent that reliably uses one connection is a stronger workshop outcome than a large agent with several unfinished capabilities.
Keep the demo prompt, exact runtime tool name, and boundary case. You will turn those three details into repeatable checks in Evals.