Limitations
No spin here - this is what codelazr actually can't do today, and where free usage gets rough around the edges. See /milestones for what it does do.
No sandbox, no shell commands
Codelazr is GitHub-native, not a terminal. It reads your repo, writes a diff, and commits it through GitHub's API - it can't run arbitrary shell commands, install system packages, or drop you into a live container to poke around. Build and test feedback comes from your own CI, not a local run.
Fix loops are bounded, not infinite
When CI fails or a review flags a gap, codelazr hands the error back to the model and lets it retry - but only up to a fixed number of attempts. If a model can't land the fix in that window, it stops rather than spinning forever, and you're back in the driver's seat.
Every task has a time limit
Long, sprawling tasks can hit a hard time cap mid-run, especially on slower free models. When that happens the task stops where it is - usually with real partial progress, not silence - rather than running indefinitely.
Free usage is real, but limited
Every user gets a bit of free usage on any model on OpenRouter, no key required - see how it works. Once that runs out, you can keep going on OpenRouter's free models at no cost.
Free models get hammered
Free OpenRouter models are shared by everyone using them, so they get hit hard - during busy periods, expect slow responses, 429s, or a queue. That's a rate limit on OpenRouter's side, not something codelazr controls. If Codelazr Cloud usage isn't landing cleanly on a free model, bring your own OpenRouter API key (BYOK) - unshared access gets you a much better response rate on the same free models, for $0. See free models, explained for more.