Dify, Langfuse, and LibreChat¶
These products are optional platforms. They are not the four AMS application modules. Add them when your use case needs agent authoring, chat, or detailed LLM tracing — and pair them with the right adapter so AMS observability stays complete.
How they relate¶
| Platform | What users do there | Why AMS cares | Typical AMS piece |
|---|---|---|---|
| Dify | Design and run agents / workflows | Agent runtime activity should be visible to AMS operators | Often self-host chart + Dify adapter |
| Langfuse | Store traces, scores, and evaluation data | Dify (and similar tools) send rich telemetry here; AMS adapters read it | Self-host chart or managed Langfuse + adapter config |
| LibreChat | Chat with models / tools | Chat usage should flow into AMS OTEL | Self-host chart + LibreChat adapter |
Dify + Langfuse + Dify adapter (common story)¶
- Dify is where builders create agents.
- Those runs emit detailed traces into Langfuse.
- The Dify adapter reads Langfuse (and/or Dify APIs as configured) and exports into OpenTelemetry.
- The observability stack stores and visualizes that data next to other AMS signals.
If you deploy Dify without Langfuse and without the adapter, agents may work inside Dify but AMS will not see that lifecycle in its telemetry plane.
LibreChat + LibreChat adapter¶
- Users chat in LibreChat.
- The LibreChat adapter collects integration telemetry.
- Data lands in AMS observability via OTEL.
LibreChat does not replace Command Center; it is an additional chat surface you may choose to run.
Charts you may use¶
| Purpose | Chart (typical) |
|---|---|
| Dify self-host | ams-helm-dify-selfhost |
| Langfuse self-host | ams-helm-langfuse-selfhost |
| LibreChat self-host | ams-helm-librechat-selfhost |
| Bridge Dify/Langfuse → AMS | ams-helm-dify-adapter |
| Bridge LibreChat → AMS | ams-helm-librechat-adapter |
Availability of self-host charts can differ by environment policy (some internal envs only deploy them in Dev/Prod). For customer release, follow the package your delivery team provides.
Practical guidance¶
- Start from the module decision (Command Center / ARE / EOBS / FortifyOps), not from Dify.
- Add Dify when the customer needs an agent authoring platform.
- Add Langfuse when you need durable traces the adapter can read.
- Add LibreChat when a LibreChat UX is in scope.
- Always install the matching adapter if AMS observability must reflect that platform.
- Keep platform secrets and model provider keys out of Git.
Suggested order when all three platforms are in scope¶
- Observability
- Langfuse (if used)
- Dify and/or LibreChat
- Matching adapters
- Confirm a test agent / chat appears in Grafana or your trace UI