Skip to main content
Context length controls how much conversation history your AI model can remember and use when generating responses. It is measured in tokens. A larger context length allows the model to consider more previous messages, while a smaller one limits how far back it can look.

​
How to adjust context length

1

Open Settings

Open the sidebar and tap Settings.
2

Find Context Length

Locate Context Length under the chat options.
3

Set a value

Use the slider to quickly select popular preset values like 256, 512, 2048, 4096, and up to 200k tokens. For custom values, type a number directly in the text field below the slider. The minimum allowed is 256 tokens.
4

Enable override

Enable the Override toggle to use your custom value instead of the provider’s default.

​
Where it applies

  • Local Server (LM Studio / Ollama): sets the context window for local chat requests.
  • OpenRouter: overrides the model’s default context length. When enabled, it also activates automatic conversation compaction to stay within the limit.
  • Custom Endpoint: applies your setting to OpenAI-compatible servers.

​
Provider defaults

When Override is disabled, LMSA uses the context length your AI provider publishes. LM Studio shows a Context Snapshot card displaying your current provider, model, and its native context limits.

​
When to adjust

  • Increase it if you want longer conversations with better memory of earlier messages, especially for complex multi-turn topics.
  • Decrease it to reduce processing time and latency, or if you run out of VRAM on a local model.
  • Use a smaller value for lightweight models or slower hardware.
  • Use a larger value for detailed analysis, coding tasks, or when you need the model to remember many prior turns.
Context length is separate from Max Output Tokens: the former controls input history, the latter controls response length.
  • Larger context lengths use more memory and may be slower on weaker hardware.
  • Not all models support the full context limit you set. Your model may truncate or ignore your override if it exceeds the model’s hard maximum.
  • Web Search and Image Generation use independent token limits separate from this setting.