Fetch and compile a versioned Langfuse prompt at runtime using labels, client-side caching, and a fallback for outage resilience

Documented steps

1. Fetch by name; the production-labelled version is returned by default when you omit label and version. Python: `prompt = langfuse.get_prompt("movie-critic")`. JS/TS: `const prompt = await langfuse.prompt.get("movie-critic")`.
2. For chat prompts, pass the type explicitly: Python `langfuse.get_prompt("movie-critic-chat", type="chat")`; JS/TS `await langfuse.prompt.get("movie-critic-chat", { type: "chat" })`.
3. To get the newest draft rather than the deployed version, request the `latest` label: `langfuse.get_prompt("movie-critic", label="latest")` (Python) or `langfuse.prompt.get("movie-critic", { label: "latest" })` (JS/TS).
4. Pin an exact version for reproducibility with `version=<int>` (Python) or `{ version: <int> }` (JS/TS). Version and label are mutually exclusive selectors; choose one.
5. Tune staleness versus request volume with an explicit cache TTL: Python `langfuse.get_prompt("movie-critic", cache_ttl_seconds=300)`; JS/TS `langfuse.prompt.get("movie-critic", { cacheTtlSeconds: 300 })`. The default TTL is 60 seconds; set 0 to disable caching in development.
6. Always pass a fallback so a cold start during a Langfuse outage cannot crash the app: Python `langfuse.get_prompt("movie-critic", fallback="Do you like {{movie}}?")`; for chat prompts pass `type="chat"` plus a fallback message array. JS/TS mirrors this with `{ fallback: ... }`.
7. After fetching, check `prompt.is_fallback` (Python) or `prompt.isFallback` (JS/TS). If true you are serving a degraded local prompt rather than the managed one, so log or alert on it.
8. Compile variables before sending to the LLM: Python `compiled = prompt.compile(criticlevel="expert", movie="Dune 2")`; JS/TS `const compiled = prompt.compile({ criticlevel: "expert", movie: "Dune 2" })`.
9. For chat prompts that need conversation history injected mid-array, define a placeholder message `{ "type": "placeholder", "name": "chat_history" }` in the prompt, then compile with both variables and the placeholder fill: `prompt.compile(criticlevel="expert", chat_history=[...])` (Python) or `prompt.compile({criticlevel:"expert"}, {chat_history:[...]})` (JS/TS).
10. Read model parameters (model, temperature, and so on) from `prompt.config` rather than hardcoding them, so they can be tuned in the Langfuse UI without a code deploy.
11. Verified against current official docs on 2026-07-29: https://langfuse.com/docs/prompt-management/get-started , https://langfuse.com/docs/prompt-management/features/caching , https://langfuse.com/docs/prompt-management/features/guaranteed-availability , https://langfuse.com/docs/prompt-management/features/message-placeholders