docs: add Streamlit integration cookbook by umutdinceryananer · Pull Request #2821 · langfuse/langfuse-docs

umutdinceryananer · 2026-04-19T22:23:36Z

What

Adds a Streamlit integration cookbook covering end-to-end Langfuse
tracing in a streaming chat app — sessions, user identification,
user-feedback scores, and live routing between OpenAI and Anthropic
with provider-tagged traces.

Cookbook notebook: cookbook/integration_streamlit.ipynb
Rendered guide: content/integrations/other/streamlit.mdx
Route entry: cookbook/_routes.json (isGuide: false)
Sidebar entry: content/integrations/other/meta.json
Logo: public/images/integrations/streamlit_icon.svg
Runnable companion: examples/streamlit-demo/

Why

Streamlit is one of the most common ways Python devs ship LLM UIs, but
there was no first-party Langfuse walkthrough. The guide mirrors the
structure of the existing Gradio cookbook so readers can jump between
the two without re-learning the layout.

Covered features

@observe decorator wrapping a generator-based streaming handler
langfuse.openai drop-in + AnthropicInstrumentor for auto-tracing
propagate_attributes(session_id, user_id, tags=[provider])
Per-tab session scoping via st.session_state
👍 / 👎 feedback wired to create_score with CATEGORICAL data type
Live OpenAI (gpt-4o-mini) / Anthropic (claude-sonnet-4-6) switch
via sidebar selector, with the provider tagged on every trace

Notes for maintainers

Runnable companion. Streamlit apps can't render inside a notebook
the way Gradio does, so the cookbook links out to examples/streamlit-demo/
as a one-command run. If the convention here is docs-only, I can move
the code inline and drop the folder.
Screenshots. "Explore data in Langfuse" reuses existing official
docs assets (/images/docs/session.png, /images/docs/users-list.png)
since the dashboard views are framework-agnostic. Happy to capture
Streamlit-specific traces and host them under
static.langfuse.com/cookbooks/streamlit/..., or link a public demo
trace, if you'd prefer.

Test plan

pnpm link-check passes for content/integrations/other/streamlit.mdx
examples/streamlit-demo/app.py runs against Langfuse Cloud; traces, sessions, users, feedback scores, and provider tags all populate
pnpm dev renders the integration page with all assets resolving

Disclaimer: Experimental PR review

Greptile Summary

This PR adds a first-party Streamlit integration cookbook covering end-to-end Langfuse tracing (sessions, user identification, user-feedback scores, multi-provider routing) along with a runnable companion app in examples/streamlit-demo/.

P1 — IndexError risk: trace_holder[0] in app.py (line 161) and the matching MDX snippet are accessed unconditionally. If stream_reply fails before its generator body executes, trace_holder is empty and the next st.rerun() will crash with IndexError. A guard (trace_holder[0] if trace_holder else None) fixes this safely.

Confidence Score: 4/5

Safe to merge after guarding the trace_holder access; everything else is documentation/style.

One P1 defect: unchecked index access on trace_holder will crash the app on any streaming failure. The two P2 findings (missing else branch in setup snippet, mid-document import) are minor docs inconsistencies. Fixing the P1 guard is a one-liner.

examples/streamlit-demo/app.py (line 161) and the matching block in content/integrations/other/streamlit.mdx

Important Files Changed

Filename	Overview
examples/streamlit-demo/app.py	Runnable Streamlit app with Langfuse tracing; potential IndexError on line 161 if trace_holder is empty after a streaming failure
content/integrations/other/streamlit.mdx	Main integration guide; setup snippet silently ignores auth failures and has a mid-document import that diverges from project conventions
cookbook/integration_streamlit.ipynb	Notebook version of the guide; code mirrors the MDX and composes cleanly into a single app.py
examples/streamlit-demo/requirements.txt	Dependencies match the install instructions in the guide; no pinned versions other than langfuse>=4.0.0
content/integrations/other/meta.json	Adds streamlit entry to the integrations sidebar in alphabetical order; looks correct
cookbook/_routes.json	Adds integration_streamlit.ipynb route with docsPath and isGuide:false; consistent with adjacent entries
examples/streamlit-demo/README.md	Clear setup and run instructions with links back to the integration guide
examples/streamlit-demo/.env.example	Template env file with placeholder values for all required keys; no secrets present
public/images/integrations/streamlit_icon.svg	SVG logo for the sidebar entry; no issues

Sequence Diagram

sequenceDiagram
    participant User as Browser Tab
    participant ST as Streamlit UI
    participant LF as Langfuse (@observe)
    participant OAI as OpenAI / Anthropic

    User->>ST: types prompt
    ST->>ST: append user message to session_state
    ST->>LF: propagate_attributes(session_id, user_id, tags)
    ST->>LF: stream_reply() [generator, @observe]
    LF->>LF: start trace, capture trace_id to trace_holder
    LF->>OAI: streaming API call (langfuse.openai or AnthropicInstrumentor)
    OAI-->>ST: stream chunks via st.write_stream
    ST->>ST: append assistant message + trace_id
    User->>ST: clicks thumbs up or thumbs down
    ST->>LF: create_score(trace_id, positive/negative, CATEGORICAL)
    LF-->>ST: flush and ack

Prompt To Fix All With AI

This is a comment left during a code review.
Path: examples/streamlit-demo/app.py
Line: 160-162

Comment:
**`trace_holder[0]` IndexError on empty list**

`trace_holder[0]` is accessed unconditionally after `st.write_stream`. If `stream_reply` raises before `trace_holder.append(...)` executes (e.g., a Langfuse initialisation error inside the `@observe` wrapper, or an exception thrown before the generator body is entered), `trace_holder` is empty and this line crashes the app. Even a safe path where `get_current_trace_id()` returns `None` would silently pass `None` to `create_score`, causing an API error on feedback.

```suggestion
        trace_id = trace_holder[0] if trace_holder else None
        st.session_state.messages.append(
            {"role": "assistant", "content": reply, "trace_id": trace_id}
        )
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: content/integrations/other/streamlit.mdx
Line: 71-77

Comment:
**Auth-failure branch missing in snippet**

The setup snippet's `init_langfuse()` only prints on success; it silently does nothing when `auth_check()` returns `False`. The full `app.py` (line 25) includes the `else` branch that tells users what to fix. The snippet is what most readers will copy first, so omitting the failure path makes misconfiguration invisible.

```suggestion
@st.cache_resource
def init_langfuse():
    client = get_client()
    if client.auth_check():
        print("Langfuse client authenticated")
    else:
        print("Langfuse authentication failed — check LANGFUSE_PUBLIC_KEY / LANGFUSE_SECRET_KEY / LANGFUSE_BASE_URL")
    atexit.register(client.flush)
    return client
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: content/integrations/other/streamlit.mdx
Line: 133

Comment:
**Inline import mid-document**

`from langfuse import observe, propagate_attributes` appears in a mid-page code block rather than at the top of the file. Readers who stitch the snippets together into one script may end up with mid-module imports. The project rule is to move all imports to the top of the module. Consider moving this line into the initial "Setup" imports block alongside the other `langfuse` imports, and noting in the prose that both names are imported there.

**Rule Used:** Move imports to the top of the module instead of p... ([source](https://app.greptile.com/review/custom-context?memory=c960fc07-9928-409f-a18b-a780cbdded12))

**Learned From**
[langfuse/langfuse-python#1387](https://github.com/langfuse/langfuse-python/pull/1387)

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "chore: remove stray .claude/settings.loc..." | Re-trigger Greptile}

Greptile also left 2 inline comments on this PR.

Context used:

Rule used - Move imports to the top of the module instead of p... (source)

Learned From
langfuse/langfuse-python#1387

Anchors examples/streamlit-demo/ so subsequent commits can add the chat app, notebook sources, and README. Ignores venv, .env, __pycache__, and .ipynb_checkpoints for local dev. Closes #12

@observe

Pins langfuse to the SDK v3 line (OpenTelemetry-based) so later commits can rely on get_client() and @observe from langfuse directly. streamlit, openai, and python-dotenv float to latest. Closes #13

.env.example lists Langfuse credentials (public key, secret key, base URL for EU vs US region selection) and the OpenAI API key needed by the baseline chat app. Users copy to .env and fill in. Closes #14

Baseline Streamlit app using st.chat_input and st.chat_message, with a plain OpenAI call against gpt-4o-mini. Chat history persists across reruns via st.session_state. No Langfuse wiring yet — subsequent commits layer in tracing, sessions, feedback. Closes #15

Covers prerequisites, venv setup, dependency install, env file creation, and the streamlit run command. Notes that Langfuse credentials stay as placeholders until later commits wire up tracing. Closes #16

Wraps get_client() + auth_check() in @st.cache_resource so the credential check runs once per session instead of on every Streamlit rerun. No tracing yet — this commit only verifies the client can authenticate against the configured Langfuse instance. Closes #17

Swap `from openai import OpenAI` for `from langfuse.openai import OpenAI`. Every chat.completions.create call now emits a trace + generation to Langfuse without any further code changes.

Extract the OpenAI call into `generate_reply` and decorate it with `@observe()`. The trace now has a `generate_reply` parent span with the auto-instrumented OpenAI generation nested as a child, giving a clean request-level grouping instead of a flat generation.

Register `client.flush` with `atexit` inside the cached initializer so any pending spans in the OpenTelemetry buffer are exported before the Streamlit process tears down. Prevents trace loss on Ctrl+C / hot-reload.

Generate a uuid4 on first page load and keep it in Streamlit's session_state so it survives reruns within the same browser tab. Sets up the identifier that will be attached to Langfuse traces in the next commit.

@observe

Pass the session_state uuid into `generate_reply` and call `langfuse.update_current_trace(session_id=...)` inside the @observe span. All traces from one browser tab now group under a single session in the Langfuse Sessions view.

@observe

Langfuse SDK v4 removed `update_current_trace` on the client. Trace-level attributes (session_id, user_id, tags, metadata) are now set via the `propagate_attributes` context manager, which must wrap span creation so inheritance works correctly. Wrap the `generate_reply` call in `propagate_attributes(session_id=...)` so the @observe span inherits the session_id at creation time.

Adds a sidebar reset button that clears the chat history and generates a fresh session_id. Lets users start a new Langfuse session on demand within the same browser tab without refreshing the page.

The app relies on v4-only APIs (`propagate_attributes`, the v4 `@observe` and `get_client` pattern). The previous `>=3.0.0` pin would allow a fresh install to resolve to v3 and break at runtime. Bumping the minimum to v4 prevents that drift.

Return the Langfuse trace_id alongside the reply from `generate_reply` and store it on the message dict in `session_state`. Sets up the hook for the thumbs feedback UI, which needs a trace_id to attach scores to.

Render 👍 / 👎 buttons under each assistant message. Clicking either posts a numeric `user-feedback` score (1 or -1) to the message's trace via `langfuse.create_score` and marks the trace as scored in session_state so the buttons don't re-appear on rerun.

…iately The new assistant message was only rendered inline with `st.markdown` and was not passing through the history loop where the thumbs buttons are drawn. Users had to send a second message before the first one showed feedback controls. Triggering `st.rerun()` after the append forces the history loop to redraw the new message with its buttons.

Use CATEGORICAL data type with "positive" / "negative" values instead of NUMERIC 1 / -1. Matches the common convention for binary feedback, renders as a clean label in the Langfuse UI (no trailing zeros) and groups naturally in score aggregations by category.

`create_score` buffers the event on the batch queue, so scores weren't appearing in the Langfuse dashboard until the process exited (when the atexit handler fired). Calling `flush` right after the score keeps the UX immediate: click thumbs, refresh Langfuse, see the score.

@observe

Convert `generate_reply` into a generator that yields delta tokens from a streaming chat completion, then render it with `st.write_stream` so users see the reply typed out rather than a sudden block. The @observe wrapper keeps the span open until the generator is exhausted, so the full streamed content is still captured as one Langfuse generation. Trace_id capture moves to a small list passed in by the caller because a generator can't return a second value.

Pass `stream_options={"include_usage": True}` so OpenAI sends a final chunk with exact prompt/completion/total token counts. The langfuse wrapper reports OpenAI's numbers instead of falling back to a local tiktoken estimate, which keeps cost attribution accurate. Guard the loop against the terminal usage chunk (which has empty `choices`) so iteration doesn't IndexError on the last frame.

Adds "Your name" text input in sidebar, stores in st.session_state["user_id"]. Defaults to "anonymous" when blank. Closes #28

@observe

Adds langfuse.update_current_trace(user_id=...) inside the @observe handler so chat traces are tagged with the active user. Closes #29

Adds anthropic and opentelemetry-instrumentation-anthropic to requirements.txt for upcoming multi-provider support. Closes #30

Wraps the Anthropic SDK with OpenTelemetry instrumentation once at app startup, guarded against Streamlit reruns via @st.cache_resource. Sets up the plumbing for Anthropic provider routing in #32. Closes #31

Sidebar dropdown selects model; routes to OpenAI or Anthropic SDK accordingly and tags traces with provider name. Closes #32

Adds the empty notebook shell with NOTEBOOK_METADATA, intro, "What is Streamlit?"/"What is Langfuse?" blockquotes, STEPS_START/STEPS_END markers, and LearnMore trailer. Steps will be filled in subsequent commits. Closes #33

Adds the empty notebook shell with NOTEBOOK_METADATA, intro, "What is Streamlit?"/"What is Langfuse?" blockquotes, STEPS_START/STEPS_END markers, and LearnMore trailer. Also registers the notebook in cookbook/_routes.json so the build script processes it. Steps will be filled in subsequent commits. Closes #33

Adds Step 1 markdown header and %pip install cell covering streamlit, langfuse, openai, anthropic, the OTel Anthropic instrumentor, and python-dotenv. Closes #34

Adds env var setup (Langfuse + OpenAI + Anthropic keys) and get_client() with auth_check() verification. Closes #35

Adds the baseline chat app — OpenAI client, st.session_state for message history, st.chat_input/chat_message UI. No Langfuse instrumentation yet. Closes #36

Layers Langfuse on the baseline: langfuse.openai drop-in for automatic LLM tracing, @observe-wrapped handler for one trace per turn, and @st.cache_resource init for the Langfuse client to survive Streamlit reruns. Closes #37

Generates a session_id with uuid, attaches it to traces via propagate_attributes, and adds a "New conversation" button that resets the message history and mints a fresh session. Closes #38

@observe

Adds 👍/👎 buttons under each assistant message wired to langfuse.create_score with data_type="CATEGORICAL". Captures trace_id during the @observe call and stores it on the message so each trace can be scored exactly once. Closes #39

Switches the handler to a streaming generator with stream=True, renders chunks via st.write_stream, and requests authoritative token usage with stream_options={"include_usage": True}. Uses a trace_holder list to surface the trace_id since generators can't return a tuple. Closes #40

Adds a "Your name" sidebar input that persists in st.session_state.user_id (defaults to "anonymous") and is attached to traces via propagate_attributes alongside the session_id. Closes #41

Adds Anthropic alongside OpenAI: AnthropicInstrumentor at startup so OpenTelemetry spans land under the same trace, a sidebar model selector, and provider-named tags via propagate_attributes for filtering in the Langfuse UI. Closes #42

Closes the cookbook with a tour of the Langfuse UI: traces, sessions, users, scores, tags, and shareable trace links. References screenshot assets at langfuse.com/images/cookbook/ integration-streamlit/ to be uploaded with the upstream PR. Closes #43

Official color mark SVG from Streamlit's brand kit (https://streamlit.io/brand), ~1.7KB, aligned with the pattern used for other integration icons in this directory. Closes #46

Adds "streamlit" to content/integrations/other/meta.json so the Streamlit page appears in the left-nav under Integrations → Other, alphabetically between promptfoo and testable-minds. Closes #47

Assistant messages carry a trace_id field in st.session_state for feedback scoring. OpenAI silently ignores the extra key but Anthropic rejects the request with: BadRequestError: messages.1.trace_id: Extra inputs are not permitted Project the history down to {role, content} inside stream_reply before handing it to either provider. Applied identically to the cookbook notebook's Step 9 cell. Closes #49

Captures referenced by integration_streamlit.ipynb Step 10: - streamlit-trace.png: stream_reply trace detail with nested anthropic.chat generation, session/user/tag metadata, and negative user-feedback score - streamlit-session.png: session view with 3 chronological generations for user umut, including a positive score - streamlit-users.png: users listing with 7 users, event counts, tokens and cost columns Closes #48

Generated by scripts/update_cookbook_docs.sh from cookbook/integration_streamlit.ipynb (routed via _routes.json with isGuide=false). NOTEBOOK_METADATA converted to YAML frontmatter, STEPS markers converted to <Steps> component, MARKDOWN_COMPONENT converted to <LearnMore /> import. Closes #44

Switch from the "Step 1..10" walkthrough to a gradio-style topical layout (Introduction, Setup, Implementation of Chat functions, Run Streamlit App, Explore data in Langfuse). Drops the repeated full-app snippets in favor of one consolidated app.py reference, cutting the rendered page from ~746 to ~415 lines. Replace the three custom retina-limited screenshots under public/images/cookbook/integration-streamlit/ with existing official Langfuse UI assets (docs/session.png, docs/users-list.png) and link to the Tracing docs + public demo project for the trace view.

The original README described the initial OpenAI-only baseline. The app has since grown to cover Langfuse tracing, sessions, user identification, streaming, user-feedback scores, and multi-provider routing — update the description, prerequisites, and how-it-works bullets to match the current state, and add a pointer back to the Streamlit integration guide as the canonical reference.

The file is a local per-user Claude Code permission cache and was tracked by accident in early branch commits. Untrack it and add the path to .gitignore so it cannot be re-committed.

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.

vercel · 2026-04-19T22:23:40Z

@Automaticare is attempting to deploy a commit to the langfuse Team on Vercel.

A member of the Team first needs to authorize it.

review-notebook-app · 2026-04-19T22:23:41Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

CLAassistant · 2026-04-19T22:23:42Z

All committers have signed the CLA.

Addresses Greptile review feedback on langfuse#2821: - P1: trace_holder[0] could IndexError if stream_reply raised before the append; fall back to trace_id = None so the UI stays alive and the scored-traces guard naturally hides the thumbs buttons on rows without a trace id. - P2: the abbreviated init_langfuse snippet in the guide only printed the success path, making misconfigured credentials silent. Mirror the else branch from the full app.py so readers copying the snippet also see the auth-failure hint.

Consolidates observe and propagate_attributes into the Setup imports rather than re-importing them mid-document. Adds a brief prose note so readers know where the names came from. Resolves Greptile P2 in both the notebook and the rendered mdx.

umutdinceryananer · 2026-05-01T15:52:35Z

Quick update: all three Greptile findings now resolved (P1 IndexError guard in d7b9bcf, P2 auth-failure branch in d7b9bcf, P2 mid-document imports in d00eff5). Ready for review when you have time — happy to capture Streamlit-specific screenshots or adjust scope if helpful.

umutdinceryananer added 30 commits April 19, 2026 00:22

chore(streamlit-demo): scaffold folder with .gitignore

c0160a2

Anchors examples/streamlit-demo/ so subsequent commits can add the chat app, notebook sources, and README. Ignores venv, .env, __pycache__, and .ipynb_checkpoints for local dev. Closes #12

chore(streamlit-demo): add Python dependencies

49c0ed7

Pins langfuse to the SDK v3 line (OpenTelemetry-based) so later commits can rely on get_client() and @observe from langfuse directly. streamlit, openai, and python-dotenv float to latest. Closes #13

chore(streamlit-demo): document required env vars

8fdc2a9

.env.example lists Langfuse credentials (public key, secret key, base URL for EU vs US region selection) and the OpenAI API key needed by the baseline chat app. Users copy to .env and fill in. Closes #14

docs(streamlit-demo): add README with run instructions

3429ea9

Covers prerequisites, venv setup, dependency install, env file creation, and the streamlit run command. Notes that Langfuse credentials stay as placeholders until later commits wire up tracing. Closes #16

feat(streamlit-demo): auto-instrument OpenAI calls via langfuse wrapper

5f578e1

Swap `from openai import OpenAI` for `from langfuse.openai import OpenAI`. Every chat.completions.create call now emits a trace + generation to Langfuse without any further code changes.

feat(streamlit-demo): flush Langfuse client on process exit

cc67881

Register `client.flush` with `atexit` inside the cached initializer so any pending spans in the OpenTelemetry buffer are exported before the Streamlit process tears down. Prevents trace loss on Ctrl+C / hot-reload.

feat(streamlit-demo): persist per-user session_id in st.session_state

5f9ddfe

Generate a uuid4 on first page load and keep it in Streamlit's session_state so it survives reruns within the same browser tab. Sets up the identifier that will be attached to Langfuse traces in the next commit.

feat(streamlit-demo): add "New conversation" sidebar button

4b5c099

Adds a sidebar reset button that clears the chat history and generates a fresh session_id. Lets users start a new Langfuse session on demand within the same browser tab without refreshing the page.

feat(streamlit-demo): capture trace_id for each assistant message

b66a43e

Return the Langfuse trace_id alongside the reply from `generate_reply` and store it on the message dict in `session_state`. Sets up the hook for the thumbs feedback UI, which needs a trace_id to attach scores to.

chore(streamlit-demo): add username sidebar input

7d36774

Adds "Your name" text input in sidebar, stores in st.session_state["user_id"]. Defaults to "anonymous" when blank. Closes #28

feat(streamlit-demo): propagate user_id to Langfuse traces

2aaf3f7

Adds langfuse.update_current_trace(user_id=...) inside the @observe handler so chat traces are tagged with the active user. Closes #29

chore(streamlit-demo): add Anthropic SDK and OTel instrumentor deps

8a7a8ac

Adds anthropic and opentelemetry-instrumentation-anthropic to requirements.txt for upcoming multi-provider support. Closes #30

chore(streamlit-demo): initialize AnthropicInstrumentor at startup

dd575bc

Wraps the Anthropic SDK with OpenTelemetry instrumentation once at app startup, guarded against Streamlit reruns via @st.cache_resource. Sets up the plumbing for Anthropic provider routing in #32. Closes #31

feat(streamlit-demo): add provider selector with Anthropic routing

df1870a

Sidebar dropdown selects model; routes to OpenAI or Anthropic SDK accordingly and tags traces with provider name. Closes #32

docs(cookbook): scaffold integration_streamlit.ipynb

b9f7d05

Adds the empty notebook shell with NOTEBOOK_METADATA, intro, "What is Streamlit?"/"What is Langfuse?" blockquotes, STEPS_START/STEPS_END markers, and LearnMore trailer. Steps will be filled in subsequent commits. Closes #33

docs(cookbook): step 1 install dependencies

05c8280

Adds Step 1 markdown header and %pip install cell covering streamlit, langfuse, openai, anthropic, the OTel Anthropic instrumentor, and python-dotenv. Closes #34

docs(cookbook): step 2 configure Langfuse SDK

12f71e0

Adds env var setup (Langfuse + OpenAI + Anthropic keys) and get_client() with auth_check() verification. Closes #35

umutdinceryananer added 16 commits April 19, 2026 19:13

docs(cookbook): step 3 minimal Streamlit chat

39eccd6

Adds the baseline chat app — OpenAI client, st.session_state for message history, st.chat_input/chat_message UI. No Langfuse instrumentation yet. Closes #36

docs(cookbook): step 4 basic tracing with @observe

f2fd6b2

Layers Langfuse on the baseline: langfuse.openai drop-in for automatic LLM tracing, @observe-wrapped handler for one trace per turn, and @st.cache_resource init for the Langfuse client to survive Streamlit reruns. Closes #37

docs(cookbook): step 5 session tracking

2c45098

Generates a session_id with uuid, attaches it to traces via propagate_attributes, and adds a "New conversation" button that resets the message history and mints a fresh session. Closes #38

docs(cookbook): step 6 user feedback scores

daec79c

Adds 👍/👎 buttons under each assistant message wired to langfuse.create_score with data_type="CATEGORICAL". Captures trace_id during the @observe call and stores it on the message so each trace can be scored exactly once. Closes #39

docs(cookbook): step 8 user identification

c9e5c1a

Adds a "Your name" sidebar input that persists in st.session_state.user_id (defaults to "anonymous") and is attached to traces via propagate_attributes alongside the session_id. Closes #41

docs(cookbook): step 9 multi-provider support

36d2b15

Adds Anthropic alongside OpenAI: AnthropicInstrumentor at startup so OpenTelemetry spans land under the same trace, a sidebar model selector, and provider-named tags via propagate_attributes for filtering in the Langfuse UI. Closes #42

chore: add Streamlit logo asset

4afc91c

Official color mark SVG from Streamlit's brand kit (https://streamlit.io/brand), ~1.7KB, aligned with the pattern used for other integration icons in this directory. Closes #46

chore: register streamlit in integrations nav

fa0209b

Adds "streamlit" to content/integrations/other/meta.json so the Streamlit page appears in the left-nav under Integrations → Other, alphabetically between promptfoo and testable-minds. Closes #47

chore: remove stray .claude/settings.local.json

f8056d8

The file is a local per-user Claude Code permission cache and was tracked by accident in early branch commits. Untrack it and add the path to .gitignore so it cannot be re-committed.

claude Bot reviewed Apr 19, 2026

View reviewed changes

dosubot Bot added the size:XL This PR changes 500-999 lines, ignoring generated files. label Apr 19, 2026

dosubot Bot added the documentation Improvements or additions to documentation label Apr 19, 2026

greptile-apps Bot reviewed Apr 19, 2026

View reviewed changes

Comment thread examples/streamlit-demo/app.py

Comment thread content/integrations/other/streamlit.mdx

umutdinceryananer added 3 commits April 20, 2026 01:30

Merge branch 'main' into docs/streamlit-cookbook

dd5f721

Merge branch 'main' into docs/streamlit-cookbook

4f53a34

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

docs: add Streamlit integration cookbook#2821

docs: add Streamlit integration cookbook#2821
umutdinceryananer wants to merge 50 commits intolangfuse:mainfrom
umutdinceryananer:docs/streamlit-cookbook

umutdinceryananer commented Apr 19, 2026 •

edited by greptile-apps Bot

Loading

Uh oh!

claude Bot left a comment

Uh oh!

vercel Bot commented Apr 19, 2026

Uh oh!

review-notebook-app Bot commented Apr 19, 2026

Uh oh!

CLAassistant commented Apr 19, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

umutdinceryananer commented May 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

umutdinceryananer commented Apr 19, 2026 • edited by greptile-apps Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What

Why

Covered features

Notes for maintainers

Test plan

Disclaimer: Experimental PR review

Greptile Summary

Confidence Score: 4/5

Important Files Changed

Sequence Diagram

Uh oh!

claude Bot left a comment

Choose a reason for hiding this comment

Claude Code Review

Uh oh!

vercel Bot commented Apr 19, 2026

Uh oh!

review-notebook-app Bot commented Apr 19, 2026

Uh oh!

CLAassistant commented Apr 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

umutdinceryananer commented May 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

umutdinceryananer commented Apr 19, 2026 •

edited by greptile-apps Bot

Loading

CLAassistant commented Apr 19, 2026 •

edited

Loading