Litmus

What the company appears to be: Litmus is a self-reported open-source contribution platform that uses AI to evaluate contributor intent before code is written. The author describes it as a "two-sided platform" where contributors connect their GitHub and are matched to issues based on their real work, then engage in a conversation with an AI assistant about their approach before submitting code.

What changed: The project description states the author built this tool after observing two problems: maintainer burnout from noise in popular projects and lack of contribution opportunities in smaller repositories. The solution is described as a pre-submission evaluation system that uses AI reasoning to determine if someone can explain their approach when challenged.

The single most important open question: Does Litmus actually function as described, or is this a prototype with limited real-world utility? The author claims it can detect deadlocks and cross-PR collisions in real repositories, but there's no evidence of actual deployment, usage metrics, or customer feedback to validate these capabilities.

The description states that Litmus is:

• A "two-sided platform" for open-source contributions
• A system that evaluates contributor intent before code exists
• A tool that uses AI to match contributors to issues and assess their understanding of the codebase
• A GitHub integration that works with issue trackers and PRs
• Built as a zero-dependency Node.js server with vanilla JavaScript frontend

The author describes it as "the litmus test for open-source contributions" - a system that evaluates whether someone can explain their approach when challenged, rather than just looking at code quality.

The description states:

• The product is positioned as solving the problem of "effort in open source being distributed terribly"
• It addresses both maintainer burnout and contributor frustration
• The author claims it's a "conversation" that happens before code exists, not after
• It's described as a tool that "catches real contributors before they write a line, filters AI slop before it reaches you"
• The positioning evolved from observing two ends of the same problem: overworked maintainers and under-served repositories

The description states:

• For contributors: People who want to contribute but don't know where to start or whether they'll be welcome
• For maintainers: Those managing large projects that are overwhelmed by noise, particularly those with CI, linters, and review bots that only look at code after it's written
• The author specifically mentions "maintainers of large projects" who are "exhausted" and "have CI, linters, review bots and static analysis"
• The target is described as people who want to contribute but have no idea where to start or whether they would even be welcome

Not evidenced. The description does not contain any information about pricing, revenue models, monetization strategies, or business model details.

The description states:

• Built with: css3, github-api, github-apps, github-webhooks, gpt-5.6, html5, javascript, json, node.js, openai-api, openai-codex, rest-api, rsa
• The whole thing is deliberately small: zero dependency Node server and vanilla JavaScript frontend
• No framework, no build step, no node_modules at all
• Uses GPT 5.6 as the engine (not just assistant)
• Every call goes through Responses API with strict JSON schema
• Repo memory system that walks git tree through GitHub API
• Selects most structurally interesting files and resolves import statements
• Scoring formula for structural risk: $R_f = \min\left(100,\ 8 d_f + 12 s_f + 12 \cdot \mathbb{1}[\text{core}] + 12 \cdot \mathbb{1}[\text{convention}] + 4 \cdot \mathbb{1}[\text{test}]\right)$
• Authenticates as GitHub App using RS256 JWT signing
• Deployed as single container on Cloud Run

Not evidenced. The description contains no information about:

• Revenue or ARR
• Customer base or adoption metrics
• Usage statistics or engagement data
• Product maturity indicators beyond the author's own account
• Any form of traction, growth, or market validation

The description states:

• Maintainers of large projects are "exhausted" and have CI, linters, review bots and static analysis
• The problem is that these existing tools look at code that has already been written
• The Jazzband Python collective shut down entirely (as an example of maintainer burnout)
• Curl shutting down a six year old bug bounty program because noise overwhelmed signal
• The author mentions "the same problem seen from two ends" - burnout in popular projects vs lack of contribution opportunities in smaller ones

The description states:

• The system is described as a prototype built for a hackathon (OpenAI 2026 hackathon)
• No evidence of actual deployment or real-world usage
• The author admits to "a three minute demo against forty to a hundred and twenty second model calls" - suggesting it's not optimized for production use
• The system is described as "deliberately small" which may indicate limited scalability
• The author mentions challenges like "getting the model to shut up" and making it stay silent when a plan is genuinely fine
• The system uses a single developer (vidhaan khare) with no team mentioned
• No evidence of any business model, monetization or revenue streams

1. What specific evidence do you have that this system actually works as described in real repositories?
2. How does the system handle edge cases where the AI cannot properly assess a plan due to ambiguity or incomplete information?
3. What is your plan for scaling beyond the current prototype and hackathon deployment?
4. How do you ensure the AI's reasoning remains accurate and relevant across different codebases and domains?
5. What are the actual technical limitations of the current implementation that would prevent production deployment?
6. How do you address the risk of false positives or negatives in the AI assessments?
7. What is your roadmap for expanding beyond the current GitHub integration and specific use cases?