#"Engine Overloaded" messages constantly even without high load

1 messages · Page 1 of 1 (latest)

south lotus
#

Bug Summary:
Kimi returns "engine overloaded" messages repeatedly during normal work sessions, causing significant workflow interruptions. The issue persists even when switching from kimi-k2-thinking to kimi-k2.

Reproduction Steps:

  1. Start a work session in OpenCode using Kimi as the engine
  2. Begin interacting with code/normal tasks
  3. Receive error messages:
  The engine is currently overloaded, please try again later [retrying in 2s attempt #3]
  The engine is currently overloaded, please try again later [retrying in 12s attempt #4]
  The engine is currently overloaded, please try again later [retrying in 26s attempt #5]

Observed behavior:

  • Messages appear in retry cycles (attempt #3, #4, #5)
  • Wait times increase progressively (2s → 12s → 26s, or 4s → 13s → 29s)
  • The process halts until retry succeeds or is abandoned

** Your Setup:**

  • Device: Desktop PC
  • OS: Ubuntu 25.10
  • OpenCode version: v1.14.48
  • Kimi model: kimi-k2 (version 2.6)

Visuals:
See attached screenshots

See This Too?
React with 🐛 if you're experiencing the same issue. This helps us spot patterns fast!

Captura_desde_2026-05-13_12-33-23.png Captura_desde_2026-05-13_12-27-45.png Captura_desde_2026-05-13_12-18-45.png
#

Update: Still receiving overload messages. Now hitting attempt #7

Captura_desde_2026-05-13_12-44-27.png
obsidian marsh
#

I started experiencing this over 24 hours ago in OpenCode. I see it in Kilo Code (VSCode) as well.

image.png
ripe moth
#

Same here, getting this a lot in OpenCode

vast crescent
#

Same here

image.png
obsidian marsh
#

Still happening.

image.png image.png
obsidian marsh
#

I found the root cause of the issue - it appears that Kimi is rate-limited and/or set to a lower priority for user agents that don't match KimiCLI/ pattern. https://github.com/anomalyco/opencode/issues/22608

I patched OpenCode to use a KimiCLI User-Agent. Testing afterwards shows that the issue is resolved and the latency and response time is significantly reduced. Good news, but I lost nearly 2 days of AI productivity due to this.

perl -i -pe 's/"User-Agent": `opencode\/\$\{InstallationVersion\}`/"User-Agent": (process.env.OPENCODE_USER_AGENT || "KimiCLI\/1.37.0")/g' opencode/packages/opencode/src/session/llm.ts
image.png
copper kelp
#

I can't use it for the fourth day in a row! I had 20 attempts today with a 2000-second wait! I couldn't stand it anymore and started using GLM again. It's unbearable, and I hope the developers will fix it soon.

obsidian marsh
copper kelp
jaunty hull
#

I can't believe I had gotten used to the slow speed...its working so fast now

hoary scaffold
# obsidian marsh I found the root cause of the issue - it appears that Kimi is rate-limited and/o...

I tested this myself and it actually works. After changing the User-Agent to KimiCLI/..., the latency and response speed improved a lot compared to the default OpenCode UA.

At least for the native Kimi for Coding provider, OpenCode should probably use the KimiCLI/ User-Agent pattern by default or expose it as an official config option. It would avoid unnecessary throttling/lower-priority routing and save users a lot of debugging time.

image.png
turbid pollen
#

Just add in opencode config file:

"kimi-for-coding": {
"models": {
"k2p6": {
"headers": {
"User-Agent": "KimiCLI/1.5"
}
}
}
}

obsidian marsh
jaunty hull