Kimi Code usage limit or context too long: match the error

A Kimi Code usage limit needs a quota or account check; a context-too-long error needs a smaller request. Read the HTTP status and message before retrying. Waiting can help a temporary overload, but does not shorten an oversized conversation. In the CLI, preserve the session before compacting it or starting a new one.

Updated and sources checked: October 9, 2026.

Match the status and error keywords

The official reference distinguishes these message fragments: 403 with 5-hour usage limit, weekly (7-day) usage limit or monthly usage limit means quota exhaustion. A 403 concurrent request limit is a risk-control limit: the reference directs users to the community guidelines and an appeal with account ID, trigger time, use case and screenshots. See Kimi Code error reference.

A 429 with too many requests or currently overloaded calls for a short wait before retrying. Overload is not personal quota exhaustion. Avoid rapid repeated requests.

A 400 message size error can show a 2097152-byte ceiling; a model token limit error can show 262144 tokens. These are the reference examples, not a universal limit for every Kimi model. Shorten history or split long input across requests. Third-party clients may wrap errors differently, so compare message keywords as well as status.

Check the quota that actually applies

New plans have a rolling five-hour window but no weekly window. Legacy plans retain the seven-day refresh from the subscription date, with no unused-quota rollover. Both groups share a monthly membership quota. CLI, editor, desktop and third-party requests count toward that shared allowance; changing clients does not create fresh quota. See Kimi Code membership benefits guide.

Check remaining quota and rate-limit status in the Kimi Code Console. Monthly usage is on the Subscription page. The CLI /usage command checks usage directly. If paid continuation is appropriate, the Subscription page has an Extra Usage module; enabling it spends that balance after subscription quota is exhausted. Its Settings can set a monthly spending cap.

/usage

Reduce CLI context without deleting session files

The current CLI saves sessions under $KIMI_CODE_HOME/sessions/, defaulting to ~/.kimi-code/sessions/. Do not manually edit those files: the documentation warns that restoration can break. See Kimi Code CLI sessions and context guide.

When idle, /new (also /clear) switches to a new session and discards the active context; /sessions (also /resume) selects saved history. /compact compresses the current history and accepts a hint about what to preserve. Check essential decisions against the saved transcript after compression.

kimi --continue resumes the latest session in the current directory; kimi --session selects saved history. Those options cannot be combined. /export-md (also /export in the terminal UI) exports readable Markdown; /export-debug-zip produces diagnostic data. Review exports for sensitive code and paths before sharing.

/compact Preserve the failing command and confirmed constraints
/export-md
/new

Older CLI help and Kimi Chat are different surfaces

The existing help-center article also documents saving history, resuming sessions and manual compression. Command details can differ from the current CLI documentation above; use the instructions for the installed client, rather than mixing terminal and web export formats. See Kimi CLI session-management help.

If the context warning is in Kimi Chat

This help page concerns Kimi Chat, not Code CLI quota. For a large initial file, split the material into smaller parts. For accumulated conversation context, summarize the confirmed facts and continue in a new conversation. Deleting earlier chat messages does not release that chat context, according to the help page. See Kimi Chat troubleshooting help.

How ShardStitch helps

ShardStitch records decisions, local changes and the next step for a fresh session. It cannot raise Kimi quota, remove a risk-control restriction or enlarge a model context window. ShardStitch pricing.