LoongSuite Mem0 Instrumentation

Mem0 Python Agent provides observability for applications that use Mem0 as a long‑term memory backend.
This document shows how to install the Mem0 instrumentation, how to run a simple example, and what telemetry data you can expect.
For details on usage and installation of LoongSuite and Jaeger, please refer to
LoongSuite Documentation.

Installing Mem0 Instrumentation

# Step 1: install LoongSuite distro
pip install loongsuite-distro

# Step 2 (Option C): install this instrumentation from PyPI
pip install loongsuite-instrumentation-mem0

# Optional dependency for threading context propagation
pip install opentelemetry-instrumentation-threading

Collect Data

Example Application

Create a simple demo.py that uses Mem0:

from mem0 import Memory

memory = Memory()
memory.add("User likes Python programming", user_id="user123")
results = memory.search("What does the user like?", user_id="user123")
print(results)

Setting Environment Variables

Configure OpenTelemetry exporters before running the example:

export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=<trace_endpoint>
export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
export OTEL_SERVICE_NAME=mem0-demo

# Enable GenAI experimental semantic conventions (required for GenAI content/event features)
export OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental

# (Optional) Capture message content – may contain sensitive data
# Must be one of: NO_CONTENT | SPAN_ONLY | EVENT_ONLY | SPAN_AND_EVENT
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=SPAN_ONLY

# (Optional) Emit GenAI events (LogRecord). Requires a LoggerProvider exporter in your app.
export OTEL_INSTRUMENTATION_GENAI_EMIT_EVENT=true

Recommended: Using loongsuite-instrument

Start your application with loongsuite-instrument (recommended in root README):

export OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=SPAN_ONLY
export OTEL_INSTRUMENTATION_GENAI_EMIT_EVENT=true

loongsuite-instrument \
    --traces_exporter console \
    python demo.py

Results

In the backend (console, Jaeger, or LoongSuite), you should see:

• Spans representing Mem0 Memory / MemoryClient calls (e.g., add, search)
• Child spans for Vector Store, Graph Store, and Reranker operations (when internal phases are enabled)
• Attributes that describe the operation, user/session identifiers, providers, and result statistics

Configuration

You can control the Mem0 instrumentation using environment variables.

Core Settings

Configuration Examples

# Enable internal phases (Vector/Graph/Reranker)
export OTEL_INSTRUMENTATION_MEM0_INNER_ENABLED=true

# Enable content capture (be careful with sensitive data)
export OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=SPAN_AND_EVENT

# Enable event emission (requires LoggerProvider exporter)
export OTEL_INSTRUMENTATION_GENAI_EMIT_EVENT=true

Semantic Conventions Status

Mem0 instrumentation ships a set of semantic attributes and span names in semconv.py that are tailored for Memory / Vector / Graph / Rerank operations. These conventions are not yet part of the upstream OpenTelemetry semantic conventions and should be treated as experimental.

Compatibility

• Python: >= 3.8, <= 3.13
• Mem0 / mem0ai: >= 1.0.0
• OpenTelemetry API: >= 1.20.0

License

Apache License 2.0

Issues & Support

If you encounter problems or have feature requests, please open an issue in the
loongsuite-python GitHub repository.

Related Resources

• Mem0 Documentation
• OpenTelemetry Python
• Gen‑AI Semantic Conventions