friendli-app 0.0.3

Friendli App - AI agent framework with CLI and SDK

AWS AI Hack Day 2025: Ready to spend a full day pushing the limits of AI?

AWS AI Hack Day Micro Conference stretches into a full-day hackathon at the AWS GenAI Loft in San Francisco giving developers the time to go deeper, collaborate longer, and actually ship what they start.

What’s happening on the ground:

⚡ Hands-on technical challenge powered by FriendliAI
🧠 A 30-minute session on scaling inference and agents
👀 An early preview of Friendli Agent shared live with the community

📍 AWS GenAI Loft, 525 Market St
🗓️ August 22, 9:30 AM – 8:00 PM PT

If you're ready to experiment, connect, and create, this is where the Bay Area's AI community will be. Register here → https://lu.ma/aws-08-22-25

QuickStart Guide

The CLI binary is available as fa (alias: friendli-app). It manages app deploy/update and basic suite operations.

1. Install (Python 3.11+)

• With pip:

pip install friendli-app

2. Authenticate

• Get a Personal Access Token: https://friendli.ai/suite/setting/tokens
• Export it or pass per-command with --token.

export FRIENDLI_TOKEN=YOUR_PAT

# Verify
fa whoami

# Or without exporting
fa --token YOUR_PAT whoami

3. Deploy an example app

Each app must contain a main.py at its root. Example apps live under examples/.

# Deploy examples/simple-app with a name, project, and optional env vars
fa deploy examples/simple-app \
  --name my-simple-app \
  --project-id <PROJECT_ID> \
  -e KEY1=VALUE1 -e KEY2=VALUE2

4. Manage apps

# List apps in a project
fa list --project-id <PROJECT_ID>

# Update source archive from a directory
fa update <APP_ID> ./path/to/app

# Restart or terminate
fa restart <APP_ID>
fa terminate <APP_ID>

Notes

• Command alias: friendli-app is identical to fa.
• Example deployment prints a link to view status in the Suite UI.

Commands

Global options

• --token: Personal access token; overrides FRIENDLI_TOKEN.
• -h, --help: Show help for the current command.

Common

• fa whoami: Show logged-in user info.
  • Usage: fa whoami or fa --token <PAT> whoami
• fa version: Show CLI version.
  • Usage: fa version

Apps

• fa deploy <APP_DIR>: Deploy an app directory.
  • Options: -n, --name <NAME>, -p, --project-id <PROJECT_ID>, -e, --env KEY=VALUE (repeatable)
  • Notes: main.py must exist at app root; ~50MB directory limit; detects pyproject.toml or requirements.txt to bundle deps.
  • Example: fa deploy examples/simple-app -n my-simple-app -p <PROJECT_ID> -e KEY1=VALUE1
• fa update <APP_ID> <APP_DIR>: Update an app’s source archive.
  • Notes: main.py required; ~50MB limit.
  • Example: fa update <APP_ID> ./my-app
• fa list --project-id <PROJECT_ID>: List apps in a project.
  • Example: fa list -p <PROJECT_ID>
• fa restart <APP_ID>: Restart an app.
  • Example: fa restart <APP_ID>
• fa terminate <APP_ID>: Terminate an app.
  • Example: fa terminate <APP_ID>

Example Apps

Each example is its own Python project. See the example’s README for setup, dependencies, and usage.

• examples/simple-app: Minimal AgentApp with sync/async callbacks and streaming. README
• examples/streaming-chat-memory: Streaming chat with persistent memory (mem0), OpenAI-compatible /v1/chat/completions. README
• examples/daily-assistant-mcp: MCP server exposing practical tools (tip calc, timezone, BMI, password). README
• examples/debug-echo: Tiny FastAPI echo service for connectivity testing. README
• examples/debug-fai: FastAPI app calling Friendli Serverless via OpenAI SDK; includes passthrough endpoint. README
• examples/langgraph-research-agent: LangGraph multi-agent research workflow with streaming. README
• examples/async-crewai-agent: CrewAI-based background task agent with progress and results endpoints. README
• examples/adk-multi-agent-research: Google ADK-style multi-agent research FastAPI service. README
• examples/autogen-dev-team: AutoGen multi-agent dev team orchestrating design→code→review. README

SDK Guide

Build lightweight HTTP agents using the SDK in friendli_app.sdk.

• Import: from friendli_app.sdk import AgentApp
• Define callbacks with @app.callback (sync, async, or generators for streaming)
• Run locally: python main.py (uses Uvicorn under the hood)
• Invoke: POST /callbacks/{callback_name} with JSON body

Example

import asyncio
from friendli_app.sdk import AgentApp

app = AgentApp()

@app.callback
def greet(name: str = "World"):
    return {"message": f"Hello, {name}!"}

@app.callback
async def greet_async(name: str = "World"):
    await asyncio.sleep(1)
    return {"message": f"Hello, {name}! (async)"}

@app.callback
def stream(n: int = 3):
    for i in range(n):
        yield {"i": i, "msg": f"chunk {i+1}/{n}"}

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=8080)

Invoke callbacks

• JSON response:

curl -s -X POST http://localhost:8080/callbacks/greet \
  -H 'Content-Type: application/json' \
  -d '{"name": "Ada"}'

• Streaming (SSE):

curl -N -X POST http://localhost:8080/callbacks/stream \
  -H 'Content-Type: application/json' \
  -H 'Accept: text/event-stream' \
  -d '{"n": 5}'

Notes

• The request body JSON is mapped directly to the callback function parameters.
• Generator or async-generator callbacks stream Server-Sent Events (text/event-stream).
• To deploy an SDK app with the CLI, ensure your project root contains main.py with an AgentApp instance.