Reporting issues to Pinchy support

When something goes wrong with your Pinchy instance — an agent gives a strange answer, a tool stops working, or a response gets cut off mid-stream — we need data to help, not just a description. Because your instance is self-hosted, we cannot reach in and look at logs. Instead, you can generate a structured diagnostics file in two clicks and share it with us.

What's in the file

The diagnostics export is a single JSON file that captures:

Up to your last 10 conversation turns with the reported agent
The model and provider name plus token usage for each turn
Tool calls (name and arguments) along with their sanitized results. When a tool's arguments were too deeply nested for the trajectory to capture in full, we recover the complete values from the audit log and tag each call with an argsSource of "audit" or "trajectory" so you can see which were enriched. Very large payloads that exceed the audit-log size limit may still be shortened.
Finish reasons for each turn (stop, tool_use, aborted, timeout, etc.)
Pinchy, OpenClaw, and openclaw-node version strings
A snapshot of the agent's configuration at export time: its configured model and provider (plus the default vision model, if one is set), the tools it's allowed to use, its template and personality preset, and a SHA-256 hash of each instruction file (SOUL.md, AGENTS.md). This tells us whether a problem came from the model you chose or from drifted instructions — without us ever seeing the instructions themselves.
Recent audit-log entries scoped to your interactions with the agent
A free-form description you write before exporting (up to 500 characters)

Secrets are filtered out before the file is generated. API keys, Bearer tokens, and any values under sensitive keys (password, apikey, authorization, and friends) are automatically redacted, and the OpenClaw sessionKey is hashed with SHA-256 so it can't be replayed. Your agent's instructions are never written into the file — only a hash of them, so we can spot when they've changed without reading your custom prompt.

Choosing which chat to export

If you have more than one conversation with an agent, the dialog lets you pick which one to export. The selector lists all of your own chats with that agent — including named chats and read-only Telegram conversations linked to your account. You can only export your own chats.

When you open the dialog from a specific conversation, that chat is preselected; from Settings, the agent's default chat is preselected. If a chat's transcript is no longer on disk, we still produce a valid file from the audit-log entries alone and flag it (scope.trajectoryMissing), so support can tell an absent transcript apart from an empty one.

For Telegram chats the export includes the conversation transcript, but the audit-log section is currently limited — audit entries are attributed to your web account, so a Telegram chat's file may ship without them. The transcript is still the primary signal for debugging.

Two ways to start

From an agent reply

If a specific answer looked wrong, you can report it straight from the chat:

Hover the assistant's message.
Open the "..." menu on the message.
Click Report issue to support.

The dialog opens with that message as the anchor. In this first version, the export still bundles your most recent turns rather than slicing precisely at that message — so the surrounding context comes along for the ride. We'll narrow this down in a future release.

From an agent's Settings

If the issue isn't tied to one reply — say the agent feels generally slow, or a tool keeps failing — start from the agent's Settings:

Open the agent, then go to its Settings → Diagnostics tab.
Click Generate diagnostics export.
Choose the chat to export in the dialog (the default chat is preselected).

The export is per-agent, so the agent is already in context here — there's no agent to pick. The same dialog appears as the per-message flow.

(Looking under general Settings → Support? It now just points you here.)

The description field is small, but the right details make a big difference:

What went wrong, in one or two sentences. "The agent stopped responding after I asked it to draft an email" is better than "broken".
The time you noticed it, including your timezone. We use this to line up your export with our own monitoring.
Steps to reproduce, if you have them. Even a rough sequence helps.

Privacy and DPA

The diagnostics file may contain conversation content, tool results, and other operational data from your Pinchy instance. You decide whether and how to share it. A data processing addendum (DPA) covering data you send us is in preparation — until it lands here, contact us before sharing diagnostics files that may include personal data of third parties.