Comp.ly

Inspiration

We've both lived the SOP problem as lab students drowning in policy PDFs, and as enterprise employees where the right process is buried three clicks deep in a wiki nobody reads. The real frustration isn't that the policies don't exist it's that they never reach the people who need them, at the moment they need them.

We wanted to fix that without forcing organizations to rip out their existing tools. The insight was simple: browser agents get the same login privileges as humans. They can use the same systems, the same UIs, the same workflows no API required, no IT ticket, no integration project. That means we can meet orgs exactly where they are and automate the grunt work, so people can focus on what actually matters.

What it does

Comp.ly turns policy documents and SOPs into live, deployable workflows in seconds. You upload a PDF, paste a URL, or crawl a policy register and the platform extracts a structured workflow with steps, roles, and deployment channels. From there you can push it to enterprise search (so employees find the right process when they search), to Slack (so it surfaces in context), or hand it to a browser agent that executes the workflow directly inside your existing tools no integration required.

How we built it

The frontend is built on Next.js 16 with React 19 and the Vercel AI SDK for streaming responses, with React Flow powering the workflow graph visualization. The backend is a FastAPI service backed by the OpenAI Agents SDK for multi-step agentic workflow generation, Pinecone for RAG-based knowledge retrieval, and SQLAlchemy over PostgreSQL. For browser execution we use Browserbase and Stagehand, which let us run Playwright agents in the cloud with full session isolation same as a human sitting at the keyboard. Document ingestion supports PDF, DOCX, and live URL crawling with fallback download flows for gated policy sites.

Challenges we ran into

Getting browser agents to reliably navigate real enterprise UIs was the hardest part ServiceNow and intranet systems are not designed with automation in mind. We also had to design a crawl pipeline that handles policy sites that require authentication or serve documents behind redirects. On the product side, the hardest design question was how to make workflow review feel fast and trustworthy rather than like a black box the step inspection UI in Opportunities went through several iterations.

Accomplishments that we're proud of

• A fully working end-to-end demo: policy URL in, ServiceNow incident out, with a real browser agent doing the work
• Enterprise search injection that surfaces the right workflow directly inside an existing intranet search page
• A clean ingestion pipeline that handles PDF, DOCX, authenticated crawls, and downloadable policy documents
• Real integrations throughout , not many mocks

What we learned

Browser agents are more capable than most people assume and the main barrier to adoption isn't the technology, it's trust and observability. Users need to see what the agent is doing and why. We also learned that the hardest part of automating compliance workflows isn't the AI step it's reliably extracting structured intent from documents written in wildly inconsistent formats.

What's next for Comp.ly

Richer workflow editing letting teams customise generated steps before deployment. Native Slack app and Teams integration so workflows surface directly in conversation. Audit logging and compliance reporting so enterprises can demonstrate that policies are actually being followed. And a marketplace of pre-built workflow templates for common compliance domains like cybersecurity incident response, onboarding, and procurement.

Built With

• agentssdk
• browserbase
• chatsdk
• cloudflare
• docker
• nextjs
• openai
• pinecone
• playwright
• postgresql
• python
• r2
• railway
• servicenow
• slack
• vercel