Slow agent responses

If your agent is taking longer than usual to reply, the most likely cause is slowness on the LLM provider's side — not MultiClaw itself. Check your provider's status page (OpenAI, Anthropic, etc.) first. If they're reporting an incident, you'll need to wait it out or switch models.

LLM provider latency

The underlying language model is hosted by a third-party provider. Their server load directly affects how fast your agent responds. MultiClaw has no control over this.

What to do: Visit your provider's status page — https://status.openai.com or https://status.anthropic.com. If the provider is healthy, try switching to a faster or lighter model in the agent's Model settings.

Large context window

Every message you send includes the full conversation history. As a conversation grows, the payload sent to the model grows too — and processing time increases with it.

What to do: Start a new conversation to reset the context. This is the most reliable way to reduce latency caused by accumulated history.

Complex multi-step task

Agents working through a multi-step plan take longer by design. Each step requires the model to reason, act, and check results before moving on. This is expected behaviour, not a bug.

You'll see the agent's progress in the chat. If it appears completely stuck — no activity for several minutes — that's a different issue from general slowness.

Slow internet connection

A weak connection between your machine and the LLM provider's API adds latency to every request. Even a small amount of packet loss compounds across a long conversation.

What to do: Switch to a faster network if possible. You can use a speed test to confirm local bandwidth isn't the bottleneck.

Too many MCP tools enabled

At the start of every turn, the agent sends the schemas for all enabled MCP tools to the model. More tools means a larger payload, which slows down the first token of each response.

What to do: Disable any MCP servers you're not actively using. Open MCP Servers from the sidebar and toggle off unused servers.