I run a 'radar' that finds free LLM endpoints and auto-adopts the good ones — behind a five-part gate so it can't adopt junk

Описание и применение

Радар автоматически обнаруживает бесплатные LLM‑эндпоинты, проверяет их по пяти строгим критериям (ответ, работа с инструментами, сохранение контекста, стабильность, превосходство над текущим) и только после прохождения всех тестов добавляет модель в стек агентских задач, избавляя от ручного поиска и скрытых сбоев.

Задача

Free LLM endpoints appear and vanish constantly; manually chasing them is inefficient and risky because many free models fail silently in agent pipelines.

Результат

The radar keeps most agent work on free or cheap models, automatically adopting better free options while preventing silent failures, and frees the author from manual monitoring.

Рабочий процесс

• 1. Scan public LLM endpoints for free tiers.
• 2. For each candidate, send a simple request to confirm responsiveness within a timeout.
• 3. Run a concrete tool‑calling test: the model must emit a correct call and incorporate the result in the next turn.
• 4. Test context handling by feeding a long prompt and verifying the full response is returned.
• 5. Perform repeated calls to assess stability and ensure consistent behavior.
• 6. Compare performance against the currently used model for the specific job; adopt only if better.
• 7. Record all metrics and audit logs; blocklist known dud models by name.

Доказательства

Only a model that clears all five gets adopted. Everything is measured on real tasks and written down, so adoption is a reproducible decision I can audit later, not a vibe.

Источник

Most of my agent work runs on free or cheap models, adopted automatically as better free options appear, with the expensive frontier models reserved for the genuinely hard slice.

Источник