PromptFoundry

Runtime prompt control for production apps. Change a prompt in the UI, see it reflected in your running app within seconds — no redeployment needed.

Install

pip install promptfoundry

Quick Start

from fastapi import FastAPI
from promptfoundry import PromptManager, aget_prompt_with_meta, log_prompt_usage

# 1. Initialize once at startup
manager = PromptManager(
    db_path="prompts.db",
    cache_ttl=5,
)

app = FastAPI()

# 2. Mount the UI
app.mount("/prompts", manager.mount_ui())

# 3. Use prompts in your routes
@app.get("/run")
async def run(text: str = "hello"):
    meta = await aget_prompt_with_meta("my_prompt")

    output = your_llm(meta["content"], text)  # your LLM call

    # 4. Log usage with the correct version — no hardcoding
    log_prompt_usage("my_prompt", meta["version_id"], input_text=text, output_text=output)

    return {"output": output}

Visit http://localhost:8000/prompts/list to manage prompts.

API Reference

aget_prompt(name) -> str

Returns the active version content for the named prompt. Use inside async functions (FastAPI routes).

from promptfoundry import aget_prompt

prompt = await aget_prompt("my_prompt")

aget_prompt_with_meta(name) -> dict

Returns both the content and the version_id of the active prompt. Preferred when you need to log usage accurately — no hardcoded IDs.

from promptfoundry import aget_prompt_with_meta

meta = await aget_prompt_with_meta("my_prompt")
# meta = {"content": "...", "version_id": 3}

output = your_llm(meta["content"], user_input)
log_prompt_usage("my_prompt", meta["version_id"], input_text=user_input, output_text=output)

get_prompt(name) -> str

Sync version. Works in plain scripts outside an event loop. Inside FastAPI routes, use aget_prompt or aget_prompt_with_meta instead.

from promptfoundry import get_prompt

prompt = get_prompt("my_prompt")

log_prompt_usage(name, version_id, input_text, output_text)

Writes a usage log entry to the prompt_logs table. Async, non-blocking — safe to call from sync or async code. Failures are silently swallowed and never propagate to the caller.

from promptfoundry import log_prompt_usage

log_prompt_usage("my_prompt", meta["version_id"], input_text=text, output_text=output)

Logs are viewable in the UI at /prompts/logs, filterable by prompt name.

UI Pages

Once mounted, the UI is available at your mount prefix (e.g. /prompts):

LLM Suggestions

Add LLM config to PromptManager and the "✨ Get AI Suggestion" button appears automatically on every edit page. Works with any OpenAI-compatible endpoint.

manager = PromptManager(
    llm_url="https://api.openai.com/v1/chat/completions",
    llm_api_key="sk-...",
    llm_model="gpt-4o",
)

To override the system prompt used for suggestions:

manager = PromptManager(
    llm_url="...",
    llm_api_key="...",
    llm_suggester_prompt="You are an expert at writing concise RAG system prompts. Return only the improved prompt.",
)

If llm_url is not set, the suggestion button and A/B test panel are hidden automatically.

Protected Mode

Require a password to set active versions or assign the prod tag.

manager = PromptManager(
    protected_mode=True,
    admin_password="your-password",
)

No sessions or tokens — a simple password check per action. Wrong password re-renders the form with an error.

Version Tagging

Versions can be tagged prod, staging, or experiment.

• Only one prod tag is active per prompt at a time — assigning it removes the tag from the previous version automatically.
• In protected mode, assigning prod or setting a version active requires the admin password.

Resilience

If the database is unreachable, get_prompt / aget_prompt serve the last cached value and log a warning. Your app never crashes due to a DB failure. If there is no cached value and the DB is down, a PromptNotFoundError is raised.

Config Reference