# 3 Open Source AI Tools Behind a 3.5-Hour Personal Agent Hackathon Cline, Composio and Parallel powered a 3.5-hour open source AI hackathon in San Francisco. Here is how the stack works and what the winners built. Source: https://zakaz-04481.shop/3-open-source-ai/ · based on the channel [Composio](https://www.youtube.com/channel/UCEuoGOJFuXUZZhprSAAK85Q) · Video: [Inside an Open Source AI Week hackathon (Cline × Parallel × Composio)](https://www.youtube.com/watch?v=grnJGKCZiqY) · 2026-10-23 ![3 Open Source AI Tools Behind a 3.5-Hour Personal Agent Hackathon](https://zakaz-04481.shop/3-open-source-ai/3-open-source-ai.webp) ## Key takeaways - Composio, Cline and Parallel ran an open source AI hackathon on October 19 at Frontier Tower in San Francisco. - Teams had 3.5 hours to build and open source a personal agent they would run every day. - Cline is the open-source coding agent used to build; Composio connects agents to 1,500+ apps; Parallel supplies live web search and research APIs. - First place went to Genie, which won an NVIDIA DGX Spark; second was The Daily Edit with $1,000 in Cline credits; third was Porchlight. - Judges were Max Paulus (Cline), Rayane Boussanni (Composio) and Karan Aggarwal (Parallel). Open source AI tooling now covers the full path from writing an agent to connecting it to real apps and giving it live web data. At the Open Source AI Week hackathon run by Cline, Composio and Parallel, teams used exactly that stack to ship and publish personal agents in 3.5 hours. The build environment is free to start with at [Cline, the open-source coding agent](https://cline.bot). ## What the open source AI hackathon asked builders to do The brief was narrow on purpose: build a personal agent you would actually want running for you every day, then open source it. The event took place on October 19 at Frontier Tower in San Francisco as part of Open Source AI Week, with three and a half hours on the clock from kickoff to demo. Judges were Max Paulus of Cline, Rayane Boussanni of Composio and Karan Aggarwal of Parallel. A short time box changes what people build. Instead of sprawling platforms, teams shipped single-purpose agents with a clear daily job, a working integration or two, and a public repository. The constraint also made the division of labor between the three tools obvious: one tool writes the code, one connects the agent to accounts and apps, one feeds it current information from the web. ## The three open source building blocks Each sponsor supplied one layer of the agent stack, and each layer is usable on its own. 1. **Cline** is an open-source coding agent that works inside the editor. It plans changes, edits files and runs commands, which is how teams went from an idea to a working agent loop within the time box. 2. **Composio** is the integration layer. It gives an agent authenticated access to more than 1,500 apps, including Gmail, Slack and GitHub, and handles the managed authentication that usually eats hours of setup. Documentation lives at https://docs.composio.dev. 3. **Parallel** provides web search and research APIs built specifically for agents, so an agent can pull live pages and structured answers instead of relying on stale training data. Documentation lives at https://docs.parallel.ai. Put together, the pattern is: an agent that can read your inbox, act in your workspace and check the current state of the web. That is a meaningfully different product from a chatbot, and it is the reason the open source angle matters here. Anyone can inspect the code, self-host the pieces and swap a component without asking a vendor. ## How a personal agent gets connected to real apps The hardest part of a useful personal agent is not the model call, it is authorization and tool access. Composio's approach is to expose app actions as callable tools with managed authentication, so the agent does not need custom OAuth plumbing per service. A builder describes the task, selects the tools, and the agent executes against the user's own accounts. Practical examples from the hackathon brief map directly onto this: an agent that triages email, an agent that posts updates to a team channel, an agent that watches a repository. Each requires read and write scopes, error handling and a way to confirm the action happened. Doing that for 1,500+ apps through one integration layer is what compresses a multi-day build into an afternoon. ## Why live web access separates agents from chatbots Parallel's contribution addresses a different failure mode: an agent that reasons well but knows nothing recent. Search and research APIs designed for agents return content an agent can act on, not a page of links a human has to skim. In a daily-use agent, that shows up as checking current prices, verifying a claim before acting, or summarizing what changed since yesterday. For open source projects, this matters twice over. The agent's behavior is auditable, and the data source is a replaceable dependency rather than a black box. A builder can point the same agent at a different search provider without rewriting the logic. ## Who won the Cline, Parallel and Composio hackathon Three projects took the podium after judging: - **First place: Genie**, which won an NVIDIA DGX Spark. The prize signals where the organizers see the bottleneck moving: local compute for running and fine-tuning models alongside the agent. - **Second place: The Daily Edit**, awarded $1,000 in Cline credits, useful for teams that want to keep building with the coding agent after the event. - **Third place: Porchlight**, rounding out a podium of small, focused personal agents rather than broad platforms. Notably, every finalist had to publish code. That is the difference between a demo day and an open source hackathon: the artifact outlives the event, and other builders can fork it. ## Links from the hackathon - Cline, the open-source coding agent: https://cline.bot - Composio documentation, integrations for 1,500+ apps: https://docs.composio.dev - Parallel documentation, web search and research APIs for agents: https://docs.parallel.ai ## What builders should take away from a 3.5-hour open source sprint The format proves that a working personal agent is now an afternoon project, not a quarter-long one, provided you pick the right three layers. Start with a coding agent to write and iterate on the loop, an integration platform for authenticated app access, and a live web API for current facts. Then publish the repository, because the open source part is what lets the next builder improve on your work instead of starting over. This breakdown of the Open Source AI Week hackathon comes from the Composio channel, which hosted the event alongside Cline and Parallel. If you want to build the same kind of agent, begin at [Cline](https://cline.bot) and wire in your first integration the same day. ## Questions & answers **What is an open source AI agent?** An open source AI agent is an agent whose code is published under a license that lets others inspect, run and modify it. In practice that means the prompt logic, tool definitions and integration code are all visible, so a builder can self-host the agent or swap the model and data sources without vendor permission. **What did the Cline, Parallel and Composio hackathon require?** Teams had three and a half hours to build a personal agent they would want running every day, and they had to open source it. The event was held on October 19 at Frontier Tower in San Francisco during Open Source AI Week, with judging by Max Paulus, Rayane Boussanni and Karan Aggarwal. **Which tools connect an AI agent to Gmail, Slack and GitHub?** Composio provides that integration layer, exposing more than 1,500 apps as callable tools with managed authentication. That removes most of the per-service OAuth work, so an agent can read email, post to Slack or act on a repository using the user's own accounts. **Who won the Open Source AI Week hackathon?** Genie took first place and an NVIDIA DGX Spark. The Daily Edit placed second and received $1,000 in Cline credits, and Porchlight came third. All finalists published their code as open source.