# Troubleshooting > The problems people actually hit, and how to fix them. Source: https://cotera.co/docs/troubleshooting --- The short list, based on what actually goes wrong in production. ## Tool calls failing **A tool that worked before now fails.** Most often an expired or revoked connection. Open **Tools**, find the provider, and reconnect. OAuth connections (Google, Slack, HubSpot) can be invalidated by password changes or workspace admin actions on the provider's side. **A tool fails intermittently under heavy use.** Provider rate limits. Scraping and search tools hit them first. Space the work out (schedules help), or reduce how many items the agent processes per run. **An agent retries the same failing call over and over.** The input is wrong — a channel that doesn't exist, a suspended account, a malformed URL. Agents don't give up on their own; stop the run, fix the input in the agent's instructions or the request, and rerun. **Slack messages not sending.** Check the channel name exists and the Cotera Slack app has been invited to it — `channel_not_found` is the most common Slack failure by far. **Scraping/extraction returning errors on specific sites.** Some sites block scraping or time out. Try the URL in a browser first; if the site works but extraction fails repeatedly, that site likely can't be scraped — route around it with a search tool instead. ## Chat and agents **The agent's answer is cut off or the run stopped.** Check your credit balance — runs stop when an org runs out of credits. Add credits from Settings → Billing. **An agent uses the wrong tool or ignores one.** Agents only see tools they've been given. Check the agent's tool list; if the right tool is there, name it explicitly in the instructions ("use Google Sheets to record results"). **A long conversation got slow or confused.** Compact it (Coco can summarize the conversation so far and continue fresh), or start a new chat and @-mention the relevant document or agent to carry over the context that matters. ## Datasets **A column run fails or hangs.** Check the warehouse connection (Snowflake/BigQuery credentials expire like any other). For LLM columns, run a handful of rows first before running thousands. *** Stuck on something not listed? Ask Coco — it can see your org's tool connections and recent errors — or reach us at .