Developer / Runs in your browser
OdinPulse: is that GitHub star growth real?
A research terminal for GitHub: growth charts, audience signals, engineering activity and an evidence-led AI research memo.
- Free, no account
- Uses keys you paste in
Scans call the GitHub API, the public GH Archive mirror and a few public package registries directly from your browser. Your GitHub token, and any AI key you add, stay in this browser and are sent only to those services, never to Gixts.
How it works
- 01
Paste a repo (owner/repo or its URL), an organisation, or a list of several targets, one per line.
- 02
Open Settings and paste a classic GitHub personal access token with no scopes; without one, the scan runs in a reduced anonymous mode inside GitHub's 60 requests an hour.
- 03
Pick Quick (30 stargazer profiles per repo) or Deep (150) and run the scan. Progress shows repos, profiles audited and API requests used.
- 04
Read each card: the verdict tier, the score, the weekly star chart with spike markers, the six factors with their evidence, and the questions to bring to the founders. Export as JSON or Markdown.
Questions
What is a classic token, and why do fine-grained tokens fail?
On github.com go to Settings, Developer settings, Personal access tokens, Tokens (classic), and generate one with no scopes ticked: it only reads public data. It raises your limit from 60 to 5,000 requests an hour. Fine-grained tokens (github_pat_...) are rejected by some public read endpoints such as user profiles and commit statistics, so the audit comes back empty; use a classic token.
Can I scan without a token?
Yes, but GitHub allows an anonymous address only 60 requests an hour. OdinPulse then audits 12 stargazer profiles per repo instead of 30 or 150 and expands an organisation to its two most-starred repos, and it still spends no more than 65% of that allowance. The star history itself comes from GH Archive and costs no GitHub requests at all.
How accurate are the percentages?
The audit inspects 12, 30 or 150 accounts and may deliberately oversample spike windows. Cohort percentages describe this sample, not the whole audience. The timeline counts archived star events, which can include repeated stars and gaps in archive coverage; the mirror can lag. Cumulative events are not the live star balance, and archive/live discrepancies do not establish purchased stars. Unavailable scored factors are excluded rather than treated as zero.
What does a red flag mean?
A reason to ask a question, not a verdict on the project. NEEDS ATTENTION means an event the scan could not match probably drove a window: a conference talk, a viral post, a launch. SUSPICIOUS and COMPOUND RISK mean one or more hard signals (a flock of look-alike accounts, a dirty spike with no event, mass-deleted stargazers) that deserve verification before the traction is taken at face value.
Where are my keys stored?
In memory while the page is open. If you tick Remember on this device they are saved in this browser's localStorage under gixts-odinpulse-keys, and the Forget keys button removes them. They are sent only to GitHub and, if you add one, to the AI provider you chose, straight from the browser; Gixts never sees them.
Is the AI brief required?
No. Verdicts, scores and flags come from the deterministic engine. Choose Anthropic, OpenAI, Gemini, or import custom multi-model providers using Responses or Chat Completions. The optional AI memo covers bull and bear cases, growth, evidence gaps and follow-ups. It receives collected evidence, not your personal notes. Custom endpoints must allow browser CORS; saved keys are masked on screen but not encrypted in localStorage.
Built with GH Archive (public event archive), ClickHouse playground (public github_events mirror).