An observability workspace for Claude Code
Record Claude Code requests and responses locally. Inspect prompts, tools and context, then use timelines, snapshot comparisons and AI analysis to investigate what changed during a run. Explore it yourself, or let an agent investigate further.
Expand a conversation into the request behind it.
Inspect the system prompt, history, tool definitions and call parameters alongside user input. When an answer changes or instructions conflict, start with what was actually sent.
Thinking is visible only when it was returned and recorded; this is not access to all internal model reasoning.
Actual software UI with isolated synthetic data.
Put main, subagent and auxiliary requests back into the run.
Title generation, safety checks, context compaction and retries can all happen during a task. Session lanes and turns make waits, dispatches and repeated calls easier to locate.
Some relationships are inferred by matching rules. Check the request before treating a connection as causal.
Actual software UI with isolated synthetic data.
Save a prompt or capture as a snapshot, keeping the material available for another look.
Back up a prompt or capture from request details and organize it in the analysis workspace. Preserve the context used in your investigation.
Compare prompts and context, including subtle changes in line breaks, zero-width characters and homoglyphs. Continue with a model discussion when useful.
A difference does not prove a cause. Two-prompt comparison and capture analysis are distinct workflows. Actual software UI with isolated synthetic data.
Snapshots and comparison (中文)Translate or explain text blocks directly in request details, without copying them between tools.
Use a text block’s translation action to read prompts, mixed-language tool results or model responses in translation.
Start an AI explanation from a text block for help understanding complex prompts, tool content or returned text.
These optional tools require an analysis model endpoint. Selected material is sent to that endpoint; check explanations against the original.
Details and reading assistance (中文)The UI and local HTTP API use the same records. For questions spanning many requests, an agent can search and aggregate, then return to individual record IDs.
Configure a model endpoint to translate or explain text, or continue a discussion around saved snapshots.
With the service running, ask the agent to read the built-in guide, discover available data and query it progressively.
Read CC Wire Analyzer’s /api/ai-guide first. Inspect today’s failures and slow requests, list the relevant record IDs, then check their call parameters. State what evidence is missing for any uncertain cause.
GET http://127.0.0.1:<port>/api/ai-guide
Use the port of the current instance. This is a prompt for an external agent, not a claim of continuous autonomous investigation and action.
Agent workflows (中文)Download the asset for your platform. Reading saved records does not require an analysis model; configure translation and AI features when you need them.
Choose and launch the Windows or macOS release asset.
Follow setup, open a new Claude Code session and work as usual.
Open details or the timeline. Stop the proxy when finished and check configuration restoration.
The local proxy forwards requests to the configured upstream and saves records locally. Translation, AI explanations and semantic analysis send relevant material to your configured analysis endpoint. Update checks and downloads also use the network.
Authentication headers are redacted, but bodies may still contain messages, file content and secrets. Inspect the material before sharing it. Website screenshots use prepared synthetic examples.
The current product targets Claude Code, with Windows and macOS apps and English, Chinese and Japanese UI. Other harnesses need protocol and behavior validation; research directions are not supported integrations.
Error groups, timelines and model interpretations support investigation. Conclusions depend on the evidence and analysis. The manual distinguishes current behavior from future design.
CC Wire Analyzer