If Claude will not load, Claude Code stops responding, or requests suddenly return errors, first determine which Anthropic surface is affected. Claude.ai, Claude Code, the API, the Console, and individual models can fail independently.
How to tell whether Claude is actually down.
| What you see | Likely cause | What to check |
|---|---|---|
| Claude.ai and Claude Code both fail | Broader Anthropic service disruption | Official status components and incident history |
| Only one model returns errors | Model-specific degraded performance | The named model incident on Claude Status |
| Claude loads but refuses new messages | Usage limit, capacity, or account issue | The exact message and plan reset information |
| Only Claude Code is frozen | CLI, project, shell, hook, or local environment issue | Terminal output, logs, git state, and a fresh session |
Recent Anthropic incidents show why the distinction matters.
Anthropic's July history includes service disruptions affecting claude.ai, Claude Code, Claude Code on the web, and related surfaces, alongside separate periods of elevated errors for individual Claude models. Those are different failure scopes and require different responses.
A provider incident calls for waiting or switching providers. A model-specific incident may be solved by selecting another available model. A local Claude Code failure may require restarting the session after preserving the repository state.
What to do when Claude is down mid-task.
- Stop repeated prompts and automatic retries.
- Record the exact Claude or Claude Code error.
- Capture git status, git diff, untracked files, and recent commits.
- Save terminal output and the last test that actually ran.
- Separate verified project facts from claims in the interrupted response.
- Wait for Anthropic recovery, start a clean Claude session, or continue in another AI tool.
Do not assume a Claude outage erased the work.
Claude can disappear while edited files, test output, commits, and local project memory remain intact. The dangerous part is not always losing the code. It is losing which decisions were made, which attempts failed, and which statements were never verified.
How ShardStitch recovers Claude Code work.
ShardStitch rebuilds a continuation packet from surviving project evidence and saved memory. It separates verified facts from inferred claims, preserves constraints and failed attempts, adds dependency-risk context, and routes the next action back to Claude Code or to another supported AI coding tool.
After-death recovery is strongest where the tool leaves usable local logs and project artifacts. Browser-only Claude.ai conversation details require prior capture or a hook, while repository state and saved project memory can still be recovered locally.