Troubleshooting

Agent Feels Slow or Stuck?

A practical checklist to diagnose lag, long replies, and "nothing is happening" moments without guessing.

One of the most common support messages is: "My agent is stuck." Most of the time, it is not actually frozen. It is waiting on one of a few normal bottlenecks.

Good news: this is usually fixable in under 5 minutes once you check the right things in the right order.

Why responses can feel slow

Bigger model: stronger models can take longer, especially on complex prompts.
Large tasks: asking for "do everything" in one message adds processing time.
Tool waits: web fetches, file reads, and shell commands can add latency.
Approval pause: the agent is waiting for your explicit approval on a command.
Network/API hiccups: temporary provider slowdown or rate limits.

Fast triage checklist

1) Check for an approval prompt first

If you see anything like "Approval Required," the agent is paused by design. Review and approve/deny so it can continue.

2) Look for progress signals

If the agent says it's reading files, searching, or running a command, it's working. Give it a bit more time before retrying.

3) Reduce task size

Split one huge request into 2-4 smaller steps. Smaller scoped prompts return faster and fail less often.

4) Try a lighter model for routine tasks

Use high-end models for strategy/quality-sensitive tasks, and cheaper/faster models for cleanup, summaries, and repetitive work.

5) Re-run with a clear expected output format

Ambiguous prompts cause longer reasoning and retries. Specify exactly what you want returned.

Prompt rewrites that speed things up

Slow / vague

Review my entire workspace and tell me everything I should improve.

Faster / focused

Review only /tutorials for broken links and obvious typos. Return: 1) file path 2) problem 3) exact suggested fix

Tip: avoid sending the same request repeatedly while one run is still in progress. That can create overlap and make things look slower.

Common confusion: “Context limit exceeded”

If you see a context-limit error, the model isn't frozen — your request payload got too large.

What causes it: huge pasted logs, long chat history, multiple large files, or asking for too many things at once.
Fast fix: start a fresh chat/session for the task and send only the minimum required context.
Better pattern: break work into stages (analyze → decide → generate output) instead of one mega prompt.
If sharing logs: send only the relevant excerpt + exact error line, not the entire file.

Common confusion: Discord shows offline or stops replying

Offline presence does not always mean your container is dead, but if your agent stops replying, run this quick check:

Confirm you are testing in the same server/channel where the app is installed.
Check channel-level permission overrides for View Channel, Send Messages, and Use Application Commands.
Try a direct @mention with a tiny prompt first before sending a long request.
If slash commands disappeared, re-check server Integrations restrictions.
If still unresponsive across channels, collect exact symptoms and escalate to #help.

Common confusion: “Environment frozen” / command layer appears dead

Sometimes the model is responsive but tool execution is not. You may see messages like "I can’t run commands right now" or notice that every command hangs/times out.

Symptom pattern: chat still works, but file/tool actions never complete (e.g., can’t run git status, can’t read files, searches stall).
Quick isolation test: send one tiny task (for example: "run a minimal status check and return only pass/fail").
If tiny tasks also hang: treat this as execution-layer trouble, not prompt quality.
First recovery step: restart the gateway/session cleanly, then retry the same tiny test before resuming larger work.
Avoid thrash: do not queue many retries while it is unhealthy; that usually worsens delay/confusion.

Important: this is different from a normal slow response. If command execution is dead, prompt rewrites alone will not fix it.

Common confusion: “Brave API key missing” during web research

Users sometimes see a message implying web research is blocked because a Brave key is missing. In many setups, built-in web search still works and this message is just a configuration mismatch or transient tool issue.

First check: ask for one tiny web search task (single query, 3-5 results) instead of a large research brief.
If tiny search works: proceed normally and keep requests focused.
If it fails repeatedly: switch to a fallback plan (manual links + targeted summaries) and report exact error text.
Avoid confusion: clarify whether your setup uses built-in access or a personal API key override.

Common confusion: “I can’t find files my agent created”

Another frequent "stuck" report is actually a file-path visibility issue, not execution failure.

Ask the agent to return the exact absolute path for each file it created.
Have it run a short directory listing in the expected folder to verify file names.
If needed, ask the agent to save output in a clearly named subfolder (for example, workspace/exports/).
If location is still unclear, request a quick "where I wrote files" report before continuing work.

When to treat it as a real issue

No response for >5 minutes on a simple prompt
Repeated tool errors with no progress
Same failure across multiple channels (Discord + dashboard)
Consistent slowdown for all tasks, even very small ones
Model replies are normal, but command/tool execution consistently hangs

What to include if you ask for help

You'll get faster support if you include:

The exact prompt you sent
Any error text or approval prompt text
Channel used (Discord, dashboard, etc.)
About how long you waited
What you already tried from this checklist

Bottom line: "slow" usually means "too broad" or "waiting on approval." Tighten scope, approve intentionally, and you'll see much faster, more reliable runs.