or-bit

The description states that or-bit is a visual workflow builder for browser automation. It allows users to:

• Drag nodes onto a canvas and connect them.
• Use triggers (Start) and actions (Open URL, Act, Observe, Agent, Send Email).
• Pass data between nodes using template tokens like {{nodeId.key}}.
• Execute workflows in real Chromium browsers via Playwright and Stagehand.
• Run durable jobs through Trigger.dev.
• Collaborate with teammates using Clerk organizations and Liveblocks.

The system is built on:

• Frontend: Next.js 16, React 19, React Flow v12
• Backend: Neon Postgres, Drizzle ORM, Trigger.dev
• AI models: Gemini (google/gemini-2.5-flash)
• Infrastructure: Railway, Sentry, Resend

Inference The product is a visual automation tool for browser-based tasks, combining AI interpretation with deterministic workflow execution and team collaboration features.

The author states that or-bit was built to address two bad options in browser automation:

1. Playwright scripts — brittle due to CSS selectors.
2. Chat-based agents — black boxes with no visibility into execution.

or-bit positions itself as a middle ground: deterministic structure and AI-driven steps, while being team-oriented.

Inference The positioning is self-reported and claims to offer a balance between control and AI empowerment. No evidence of market feedback or customer validation is provided.

The author states that or-bit is designed for teams, not individuals, with multi-tenancy built in via Clerk organizations.

Inference The target customer is likely B2B teams working on repetitive browser tasks (e.g., data scraping, competitor monitoring, form filling). No evidence of specific verticals or use cases beyond the author’s own stated needs.

No pricing information, business model, or monetization strategy is described. The project is a hackathon submission with no indication of revenue, customers, or paid features.

Not evidenced

The system uses:

• Chromium browser automation via Playwright and Stagehand
• AI models (Gemini) for interpretation of natural language actions
• Durable execution via Trigger.dev
• Multi-tenancy via Clerk and organization scoping
• Real-time collaboration with Liveblocks
• Session replay using WebM

Architectural decisions include:

• One browser session per run
• Template tokens as data contract (not typed ports)
• Execution contract persisted, not UI state
• Topological sort for deterministic execution
• SSRF guards and concurrency limits

Inference The technical stack and architecture suggest a well-thought-out system, with attention to security, performance, and usability. However, no evidence of production deployment or scaling beyond the hackathon.

The project is described as a hackathon submission (OpenAI 2026). No evidence of:

• Revenue
• Customers
• Users
• Product-market fit
• Adoption metrics
• Prior versions or iterations

Not evidenced

No competitive analysis or comparison to existing tools is provided. The author only mentions two extremes:

1. Playwright scripts
2. Chat-based agents

Not evidenced

• Unverified claims: All features and functionality are self-reported.
• No traction or validation: No evidence of usage beyond the hackathon.
• Security complexity: The system handles untrusted input, which is a high-risk area. While guards are described, no independent review or testing is mentioned.
• AI dependency: Reliance on a single AI model (Gemini) may limit robustness or scalability.
• Limited scope: No evidence of advanced features like scheduling, loops, or shared templates beyond stated future plans.

1. What was the actual user feedback during the hackathon?
2. Has anyone outside the team used this tool in a real-world setting?
3. How do you plan to scale browser sessions and execution limits for enterprise use?
4. Are there any known issues with AI hallucinations or model limitations in browser automation?
5. What is the current state of the product beyond the hackathon submission?