mirror of https://github.com/langchain-ai/langsmith-trace-scripts.git synced 2026-07-01 20:04:03 -04:00

Files

T

Stephen Chu 5b31b79331 Improve PII scrubbing: Use recursive field name matching

- Changed scrub_trace.sh to match field names recursively at any depth
- Works with arrays and nested objects (e.g., "content" finds all content fields)
- Simpler interface: just field names instead of dotted paths
- Safer for PII: catches sensitive data in unexpected locations
- Updated README with new usage examples and field list

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

2025-12-23 18:42:56 -05:00

3.0 KiB

Raw Blame History

LangSmith Trace Export & PII Scrubbing

Simple bash scripts to extract, scrub, and upload LangSmith traces.

Quick Start

Extract and Scrub PII

Customers extract their trace and scrub sensitive data before sending:

# 1. Extract trace
export LANGSMITH_API_KEY='your-api-key'
./extract_trace.sh 00000000-0000-0000-f319-b36446ca3f23

# 2. Scrub PII (recursively redacts field names)
./scrub_trace.sh trace_00000000-0000-0000-f319-b36446ca3f23.json "content,email"

# 3. Review scrubbed file manually

# 4. Send trace_00000000-0000-0000-f319-b36446ca3f23.scrubbed.json to support

Upload

LangChain team uploads the scrubbed trace:

export LANGSMITH_API_KEY='your-api-key'
./upload_trace.sh customer_trace.scrubbed.json "customer-issue-1234"

Requirements

bash
curl
jq (install: brew install jq or apt-get install jq)

Scripts

`extract_trace.sh`

Extract a trace by ID.

./extract_trace.sh <trace_id> [output_file]

Output: trace_<id>.json

`scrub_trace.sh`

Redact PII fields from trace using recursive field name matching.

./scrub_trace.sh <trace_file> "<field1>,<field2>,..."

Output: <trace_file>.scrubbed.json

Common fields to redact:

content - Message content (finds all content fields)
email - Email addresses
messages - Entire message arrays
query - Search queries
text - Generated text
session_id - Session IDs
user_id - User IDs
api_key - API keys

Recursive matching: Field names are matched at any depth in the JSON structure, including inside arrays and nested objects. For example, specifying content will redact all fields named content anywhere in the trace.

`upload_trace.sh`

Upload scrubbed trace to LangSmith project.

./upload_trace.sh <trace_file> <project_name>

Complete Example

Extract and Scrub PII:

export LANGSMITH_API_KEY='lsv2_pt_...'

# Extract
./extract_trace.sh a1b2c3d4-5678-90ab-cdef-1234567890ab

# Scrub
./scrub_trace.sh trace_a1b2c3d4-5678-90ab-cdef-1234567890ab.json \
  "content,email,session_id"

# Review and send trace_a1b2c3d4-5678-90ab-cdef-1234567890ab.scrubbed.json to support

Upload:

export LANGSMITH_API_KEY='lsv2_pt_...'

./upload_trace.sh trace_a1b2c3d4-5678-90ab-cdef-1234567890ab.scrubbed.json \
  "customer-acme-issue-5678"

Help

Each script has a --help flag:

./extract_trace.sh --help
./scrub_trace.sh --help
./upload_trace.sh --help

EU Region

Set LANGSMITH_ENDPOINT for EU:

export LANGSMITH_ENDPOINT='https://eu.api.smith.langchain.com'

Troubleshooting

"jq required"

brew install jq  # macOS
apt-get install jq  # Linux

"No runs found"

Trace ID doesn't exist
Trace not fully ingested (wait a few seconds)
Wrong API key

"Failed to create project"

API key invalid or expired
No write permissions

Important

Customers must manually review scrubbed files before sending to ensure all PII is removed.

3.0 KiB Raw Blame History