Optimizer
An automatic, auditable process improves GitHub metadata of the owner's public repositories. This page shows exactly how it works and what it has changed.
What it does
- Runs twice a week and picks up to 10 public repositories, prioritising weak metadata, then traffic. A repository is revisited at most every 21 days.
- Descriptions are replaced only when the current one is empty, too short or cut off.
- Topics are only added (up to 12) when the README supports them; only clear junk such as one-letter words is removed.
- README text changes happen only for hand-built tool repositories, as a pull request that edits one marked block. It is merged automatically after 3 days unless the owner closes it; closing it stops further proposals for 90 days.
- Never touched: private repositories, forks, archived repositories, the owner's profile repository and repositories that generate their own README.
How AI is used
Text is drafted by open-weight models through free tiers of several providers with automatic fallback (see below). Only the public README, description, topics and file names of a repository are sent. Every answer must pass strict checks: topics must be grounded in the README, descriptions may not contain links or hype, FAQ answers may not contain numbers missing from the README, and a second model fact-checks the result. If no model is available or a check fails, a conservative rule-based fallback is used instead.
Safety net
Metadata changes are logged with before and after values. After 14 days the dashboard compares views before and after; if views fell by more than 30% and the repository had at least 20 views, the change is rolled back automatically. With less traffic the change is kept because there is not enough data to judge.
AI providers
Last run: 2026-11-19 03:49 UTC.
| Provider | Models | Configured | Successful | Failed | Avg latency | Today |
|---|---|---|---|---|---|---|
| groq | open-weight | no key | 0 | 0 | 0 ms | 0/800 |
| cerebras | open-weight | no key | 0 | 0 | 0 ms | 0/800 |
| github-models | open-weight | yes | 0 | 23 | 0 ms | 1/100 |
| ovhcloud | open-weight | yes | 11 | 18 | 3668 ms | 2/80 |
| pollinations | open-weight | yes | 5 | 1 | 11330 ms | 0/40 |
| openrouter | open-weight | no key | 0 | 0 | 0 ms | 0/40 |
| huggingface | open-weight | no key | 0 | 0 | 0 ms | 0/100 |
| mistral | open-weight | no key | 0 | 0 | 0 ms | 0/200 |
| cloudflare | open-weight | no key | 0 | 0 | 0 ms | 0/150 |
| gemini | closed (disabled) | no key | 0 | 0 | 0 ms | 0/200 |
Recent changes
| Date | Repository | Type | Changed | Source | Status |
|---|---|---|---|---|---|
| 2026-11-19 | musfiraai-mcp-server | product | +7 topics | ovhcloud:gpt-oss-120b | applied |
| 2026-11-19 | musfira-ai-psa-deepseek-v41-flash-habitually-exfiltrates-api-keys-it-is | generated | description | ovhcloud | applied |
| 2026-11-19 | musfira-ai-fable-5gguf-when | generated | description, +0 topics, -1 junk | ovhcloud | applied |
| 2026-11-19 | talashly | product | +8 topics, README pending | ovhcloud | applied |
| 2026-11-19 | talashly-monitor | product | description, +8 topics, README pending | ovhcloud | applied |
| 2026-11-19 | musfiraai-mcp-server | product | +5 topics, README pending | ovhcloud | applied |
| 2026-11-19 | musfiraai-mcp | product | description, +6 topics, README pending | ovhcloud | applied |
See also the methodology and status page.