Ritsu is slow or laggy

If Ritsu feels slow, the cause is usually one of: peak-hour load on the AI provider, a too-large PoK / source context, browser tab pressure, or your network. Here's the triage.

Quick checks

Peak hours

AI provider latency spikes during US/EU office hours (roughly 14:00-22:00 UTC). A response that streams in 8 seconds at 03:00 UTC might take 30 seconds at peak. Not a Ritsu bug — model providers throttle. Workarounds:

• Wait it out — peak passes within hours
• For Pro users: priority routing kicks in, you'll feel less slowdown but still some

Large context

If your PoK is built on a 500-page source and you ask /explain to summarise the whole thing, Ritsu has to feed all of it to the model — which takes time + tokens.

Symptom: first response after switching to a new PoK is slow; subsequent responses are fast.

Workarounds:

• Use smaller, more focused PoKs (a chapter, not the whole book)
• Use /search first to find relevant sections, then /explain scoped to those
• Run /summarize once to get a fast overview, then dive into specific parts

Browser tabs

If you have 30 tabs open, your browser may be context-switching aggressively. Streaming connections compete for CPU.

Workaround: close tabs you're not using. Or use a fresh browser window for Ritsu.

Slow network

Streaming connections are sensitive to network jitter. A 4G connection on a moving train will feel choppy even if total throughput is fine.

Workaround: switch to Wi-Fi if you can. If the issue persists on stable Wi-Fi, see Chat won't load or respond.

Specific symptoms

Streaming starts fast, then slows mid-response

Usually a network jitter issue. The first chunks come through fast (in a buffer), then real-time pace catches up. Refresh and retry on a more stable network.

Always slow regardless of time of day

Could be your network. Run a speed test (fast.com). If <2 Mbps stable, that's the problem.

If your network is fast but Ritsu is consistently slow, email support@ritsu.ai with:

• Your timezone + a few timestamps when it was slow
• Speed-test result
• The PoK size if known (bigger PoK = bigger context)

Quiz / flashcard generation is slow

Generation tasks are heavier than chat — they involve more model passes. Expect 10-30 seconds for a 5-question quiz, 20-60 seconds for a 10-card flashcard deck. If it's much longer than that, retry — sometimes the first generation hits a slow path that doesn't repeat.

Saving a deck is slow

Network upload of the rendered deck. Usually <2 seconds. If it spins forever, refresh and retry. Saved progress isn't lost.

When to upgrade

Pro plan gets:

• Priority model routing (faster on average, especially at peak)
• Higher per-request token limits (bigger context, fewer "context too large" errors)
• Unlimited daily commands

If you're consistently at the Free plan's edges, Pro pays for itself in time saved.

Still slow?

Email support@ritsu.ai with timestamps + a sample slow query. We can correlate to provider-side latency in our metrics.