Hackathon · Open Source AI Week
Build your own open personal agent
Cline × Parallel × Composio
Build an agent that does something useful in the tools you use every day.
Bring a laptop and a problem you want to solve.
01 / Build brief
The prompt
Build an open-source personal agent that turns a task into an authenticated action in a real app.
Define the input and the expected output. Have the agent gather context, select the tools it needs, and execute the workflow against your connected accounts. Use web research where useful. The result should be visible in the target app.
- Issue description → web research → GitHub ticket
- Meeting notes → action items → tasks in Linear
- Research question → cited findings → a document update
Use Cline to build, Parallel for web research, and Composio for authenticated access to apps and tool execution. Choose the tools your workflow needs and explain how you used them. Work solo or form a team.
Open-source your project and share it with the community. We will finish with demos and prizes for the top three projects.
02 / Demo protocol
What to demo
Demo one end-to-end run. Make the input, tool execution, and resulting app state inspectable.
01 / Input
State the user's task, show the input, and define the expected output. Explain which apps and accounts are involved.
02 / Execution
Run the agent. Show the context it retrieves and the tool calls it executes. Explain how authentication is scoped to the user and where approval is required.
03 / Output
Verify the result in the target app. Identify any mocked, failed, or incomplete steps and the current limitations. Keep the demo focused on behavior you can demonstrate.
04 / Repository
Share the public repo and license. Document dependencies, environment variable names, account setup, the run command, and a sample input. Distinguish reused code from your work.
A CLI demo is welcome. Use test accounts or safe sample data. Keep API keys, tokens, and private information out of your repo and presentation.
03 / Evaluation
Judging
Four criteria. Equal weight.
Each criterion is scored from 1 to 5, for a total out of 20.
Judges use the combined score to select the top three projects. Show the actual execution and resulting app state. A CLI, notebook, or graphical app is valid; UI polish is not a separate criterion.
- Usefulness
- Show a real task and who benefits. Demonstrate how the agent saves time or improves the result compared with doing the task by hand.
- Working execution
- Run the full path from input to authenticated tool calls to an observable app result. Explain account scoping, approval boundaries, and any failed or simulated steps.
- Originality
- Show a thoughtful approach to the problem or a meaningful improvement on an existing workflow. Explain your contribution and what you reused.
- Open-source reproducibility
- Provide a public repo, an open-source license, dependencies, environment variable names, account setup, and a sample input. Another developer should be able to reproduce the run with their own credentials.
| Score | Evidence |
|---|---|
| 1 | Little evidence for this criterion, or an idea that has not been demonstrated. |
| 3 | Credible evidence, with important gaps or limitations still unresolved. |
| 5 | Convincing evidence that fully meets this criterion. |
Scores of 2 and 4 fall between these descriptions.
04 / Build stack
Build resources
Cline
Build with a coding agent
Use Cline to implement your agent, connect APIs, and iterate on the workflow.
Cline docsParallel
Search and research APIs
Retrieve current web context for your agent through Parallel's search and research APIs.
Parallel docsComposio
Auth and tool execution
Connect user accounts, discover app tools, and execute authenticated actions with Composio.
Composio docsCome with a problem. Leave with an agent.
Free to attend. Solo builders and teams welcome.