docs(integrations): add NoPII gateway

[evdevlead]

Summary

Adds NoPII to the gateways section. NoPII is a hosted privacy proxy that exposes OpenAI- and Anthropic-compatible endpoints; it tokenizes PII in outbound prompts before they reach the LLM, then restores the original values in the response.

NoPII has built-in Langfuse support: when enabled in the NoPII admin console, every proxied request emits server-side spans (sanitize, llm-call, desanitize) directly to the user's Langfuse project. Combined with the W3C traceparent header, client-side application traces and NoPII's server-side traces appear together as a single end-to-end trace in Langfuse.

Files

• content/integrations/gateways/nopii.mdx (new) — integration page covering OpenAI and Anthropic via Langfuse's @observe() decorators with traceparent propagation.
• content/integrations/gateways/meta.json — adds "nopii" to the alphabetical pages list.
• public/images/integrations/nopii_icon.svg (new) — 960-byte square logo.

Verification

All code examples on the page are lifted verbatim from runnable examples in the NoPII repo and have been validated end-to-end:

• OpenAI: https://github.com/Enigma-Vault/NoPII/tree/main/examples/langfuse
• Anthropic: https://github.com/Enigma-Vault/NoPII/tree/main/examples/langfuse-anthropic

PR prepared with the assistance of Claude.

Greptile Summary

This PR adds a new NoPII gateway integration page to the Langfuse docs, demonstrating how to route LLM calls through NoPII's PII-stripping proxy while linking client-side @observe() traces to NoPII's server-side spans via the W3C traceparent header.

• nopii.mdx: New integration page with two end-to-end examples (OpenAI and Anthropic) that correctly use langfuse.get_current_trace_id() / langfuse.get_current_observation_id() to build a valid W3C traceparent; both IDs are W3C Trace Context–compliant (32-char and 16-char hex strings respectively), so NoPII's sanitize/llm-call/desanitize spans land in the correct trace.
• meta.json: Inserts "nopii" in its correct alphabetical position in the gateways sidebar.
• nopii_icon.svg: Adds a lightweight (960-byte) vector logo for the integration card.

Confidence Score: 5/5

Safe to merge — this is a documentation-only addition with no application logic changes.

The previously raised concerns about random trace/span IDs have been fully addressed: the examples now use langfuse.get_current_trace_id() (32-char hex, W3C trace-id compliant) and langfuse.get_current_observation_id() (16-char hex, W3C parent-id compliant), producing a correctly formed traceparent. All Langfuse SDK calls (update_current_generation, get_current_trace_id, get_current_observation_id) are documented v3 SDK methods. The meta.json ordering is alphabetically correct, and the SVG logo is well-formed.

No files require special attention.

Sequence Diagram

sequenceDiagram
    participant App as Client App
    participant LF as Langfuse SDK (@observe)
    participant NoPII as NoPII Proxy
    participant LLM as LLM Provider (OpenAI/Anthropic)
    participant LFBE as Langfuse Backend

    App->>LF: "customer_lookup(prompt) [@observe]"
    LF->>LF: create trace span
    App->>LF: "call_llm(prompt) [@observe as_type=generation]"
    LF->>LF: create generation span
    LF-->>App: trace_id (32 hex), span_id (16 hex)
    App->>App: "build traceparent 00-{trace_id}-{span_id}-01"
    App->>NoPII: POST /chat/completions + traceparent header
    NoPII->>NoPII: sanitize PII to vault tokens
    NoPII->>LLM: POST sanitized prompt (no raw PII)
    LLM-->>NoPII: response (tokenized)
    NoPII->>NoPII: desanitize to restore PII
    NoPII-->>App: response (original PII restored)
    NoPII->>LFBE: server-side spans (sanitize, llm-call, desanitize) parented to traceparent
    App->>LFBE: "client-side spans via @observe flush"
    Note over LFBE: Single unified trace in Langfuse UI

Loading

_{Reviews (3): Last reviewed commit: "docs(integrations): fix nopii trace prop..." | Re-trigger Greptile}

[vercel]

@ev-engineering is attempting to deploy a commit to the langfuse Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d570d287a0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[evdevlead]

Thanks for the catch, @codex. Confirmed the issue was real: the original code generated fresh trace_id and span_id values via uuid.uuid4 instead of propagating the active Langfuse context, so NoPII's server-side spans would have landed under a different trace.

Both examples now pull the active trace and observation IDs via langfuse.get_current_trace_id() and langfuse.get_current_observation_id() and forward those as the W3C traceparent. Fix in 6891bc17. The matching upstream examples in NoPII-Examples were updated in parallel.

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.

[evdevlead]

The P1 findings raised by Codex and Greptile (random trace_id/span_id instead of propagating Langfuse's active trace context) were addressed in commit 6891bc1. Both the OpenAI and Anthropic examples now read the active trace_id and observation_id from Langfuse via langfuse.get_current_trace_id() and langfuse.get_current_observation_id(), so NoPII's server-side spans correctly attach to the same trace.

@greptileai please re-review on the latest commit.