The reason we built Recourse is because we heard many anecdotes of friends in Isla Vista who were stuck in disputes with landlords. Many—including ourselves—felt the issues were too small for a lawyer. The reality, however, is that these issues are too meaningful to ignore: withheld security deposits, rent that never returned, mold that was never tended to. After hearing these anecdotes, we took the time to actually learn what qualifies to take place in small claims court in California — how filing works, what the dollar limits are, and how daunting the process becomes when there is no professional to walk you through it. 
The pattern always remained in that people knew something wasn’t right fair, but they didn’t know where to begin. Small claims court is designed for self-representation, yet most individuals never file as the process is intimidating. We want to be the medium that is missing. We want to be a resource that helps ordinary people actually pursue small claims and get adequate compensation for their troubles. 
From a technical standpoint, we structured the agent as a pipeline: conversational intake, viability check, routing against a curated California knowledge base, then artifact generation. On the frontend, we began with a form to create a case, moved to a full-width chat, and then added a sidebar for past conversations like other AI tools. On the agent side, we built skills so the model would not have to guess legal details. That included sandbox tooling, file search, a PDF skill for uploads, and an SC-100 skill to turn messy intake into official form fields. We also built synthetic demo cases with landlord deposit disputes, incomplete stories, and realistic uploads, because the demo only works if it handles how people actually talk.
Setting up the sandbox took longer than we expected. We had to get dependencies, defaults, and tool execution working so the agent could run skills instead of just describing them. Scraping and organizing California-specific knowledge was also harder than we thought, since a wrong answer delivered confidently is worse than no answer at all. Getting the product right on UI was an ongoing process too. Each iteration taught us the goal was not to look like legal software, but to feel like talking to someone who actually read the court website. 
That is what we hope Recourse becomes. Free help for people who know something is wrong but do not know where to start, so they can stop guessing and finally get what they are owed. 
Share this project:

Updates