Agently 4 🚀

Build production‑grade AI apps faster, with stable outputs and maintainable workflows.

English Introduction | 中文介绍

---

🔥 Latest Docs | 🚀 5‑minute Quickstart | 💡 Core Features

---

📚 Quick Links

• Docs (EN): https://agently.tech/docs
• Docs (中文): https://agently.cn/docs
• Agent Systems Playbook (EN): https://agently.tech/docs/en/agent-systems/overview.html
• Agent Systems Playbook (中文): https://agently.cn/docs/agent-systems/overview.html
• Coding Agent Guide (EN): https://agently.tech/docs/en/agent-docs.html
• Coding Agent Guide (中文): https://agently.cn/docs/agent-docs.html
• Agent Docs Pack: https://agently.cn/docs/agent_docs.zip

🤔 Why Agently?

Many GenAI POCs fail in production not because models are weak, but because engineering control is missing:

Common challenge	How Agently helps
Output schema drifts, JSON parsing fails	Contract‑first output control with output() + ensure_keys
Workflows get complex and hard to maintain	TriggerFlow orchestration with to / if / match / batch / for_each
Multi‑turn state becomes unstable	Session & Memo with memory, summaries, and persistence strategies
Tool calls are hard to audit	Tool logs via extra.tool_logs
Switching models is expensive	OpenAICompatible unified model settings

Agently turns LLM uncertainty into a stable, testable, maintainable engineering system.

✨ Core Features

1) 📝 Contract‑first Output Control

Define the structure with output(), enforce critical keys with ensure_keys.

result = (
    agent
    .input("Analyze user feedback")
    .output({
        "sentiment": (str, "positive/neutral/negative"),
        "key_issues": [(str, "issue summary")],
        "priority": (int, "1-5, 5 is highest")
    })
    .start(ensure_keys=["sentiment", "key_issues[*]"])
)

2) ⚡ Structured Streaming (Instant)

Consume structured fields as they are generated.

response = (
    agent
    .input("Explain recursion and give 2 tips")
    .output({"definition": (str, "one sentence"), "tips": [(str, "tip")]})
    .get_response()
)

for msg in response.get_generator(type="instant"):
    if msg.path == "definition" and msg.delta:
        ui.update_definition(msg.delta)
    if msg.wildcard_path == "tips[*]" and msg.delta:
        ui.add_tip(msg.delta)

3) 🧩 TriggerFlow Orchestration

Readable, testable workflows with branching and concurrency.

(
    flow.to(handle_request)
    .if_condition(lambda d: d.value["type"] == "query")
    .to(handle_query)
    .elif_condition(lambda d: d.value["type"] == "order")
    .to(check_inventory)
    .to(create_order)
    .end_condition()
)

4) 🧠 Session & Memo (Multi‑turn Memory)

Quick / Lite / Memo modes with summaries and persistence strategies.

from agently import Agently
from agently.core import Session

agent = Agently.create_agent()
session = Session(agent=agent).configure(
    mode="memo",
    limit={"chars": 6000, "messages": 12},
    every_n_turns=2,
)
agent.attach_session(session)

5) 🔧 Tool Calls + Logs

Tool selection and usage are logged in extra.tool_logs.

@agent.tool_func
def add(a: int, b: int) -> int:
    return a + b

response = agent.input("12+34=?").use_tool(add).get_response()
full = response.get_data(type="all")
print(full["extra"]["tool_logs"])

6) 🌐 Unified Model Settings (OpenAICompatible)

One config for multiple providers and local models.

from agently import Agently

Agently.set_settings(
    "OpenAICompatible",
    {
        "base_url": "https://api.deepseek.com/v1",
        "model": "deepseek-chat",
        "auth": "DEEPSEEK_API_KEY",
    },
)

🚀 Quickstart

Install

pip install -U agently

Requirements: Python >= 3.10, recommended Agently >= 4.0.7.2

5‑minute example

1. Structured output

from agently import Agently

agent = Agently.create_agent()

result = (
    agent.input("Introduce Python in one sentence and list 2 advantages")
    .output({
        "introduction": (str, "one sentence"),
        "advantages": [(str, "advantage")]
    })
    .start(ensure_keys=["introduction", "advantages[*]"])
)

print(result)

2. Workflow routing

from agently import TriggerFlow, TriggerFlowEventData

flow = TriggerFlow()

@flow.chunk
def classify_intent(data: TriggerFlowEventData):
    text = data.value
    if "price" in text:
        return "price_query"
    if "feature" in text:
        return "feature_query"
    if "buy" in text:
        return "purchase"
    return "other"

@flow.chunk
def handle_price(_: TriggerFlowEventData):
    return {"response": "Pricing depends on the plan..."}

@flow.chunk
def handle_feature(_: TriggerFlowEventData):
    return {"response": "Our product supports..."}

(
    flow.to(classify_intent)
    .match()
    .case("price_query")
    .to(handle_price)
    .case("feature_query")
    .to(handle_feature)
    .case_else()
    .to(lambda d: {"response": "What would you like to know?"})
    .end_match()
    .end()
)

print(flow.start("How much does it cost?"))

✅ Is Your App Production‑Ready? — Release Readiness Checklist

Based on teams shipping real projects with Agently, this production readiness checklist helps reduce common risks before release.

Area	Check	Recommended Practice
📝 Output Stability	Are key interfaces stable?	Define schemas with output() and lock critical fields with ensure_keys.
⚡ Real‑time UX	Need updates while generating?	Consume type="instant" structured streaming events.
🔍 Observability	Tool calls auditable?	Inspect extra.tool_logs for full arguments and results.
🧩 Workflow Robustness	Complex flows fully tested?	Unit test each TriggerFlow branch and concurrency limit with expected outputs.
🧠 Memory & Context	Multi‑turn experience consistent?	Define Session/Memo summary, trimming, and persistence policies.
📄 Prompt Management	Can logic evolve safely?	Version and configure prompts to keep changes traceable.
🌐 Model Strategy	Can you switch or downgrade models?	Centralize settings with OpenAICompatible for fast provider switching.
🚀 Performance & Scale	Can it handle concurrency?	Validate async performance in real web‑service scenarios.
🧪 Quality Assurance	Regression tests complete?	Create fixed inputs with expected outputs for core scenarios.

📈 Who Uses Agently to Solve Real Problems?

"Agently helped us turn evaluation rules into executable workflows and keep key scoring accuracy at 75%+, significantly improving bid‑evaluation efficiency." — Project lead at a large energy SOE

"Agently enabled a closed loop from clarification to query planning to rendering, reaching 90%+ first‑response accuracy and stable production performance." — Data lead at a large energy group

"Agently’s orchestration and session capabilities let us ship a teaching assistant for course management and Q&A quickly, with continuous iteration." — Project lead at a university teaching‑assistant initiative

Your project can be next.
📢 Share your case on GitHub Discussions →

❓ FAQ

Q: How is Agently different from LangChain or LlamaIndex?
A: Agently is built for production. It focuses on stable interfaces (contract‑first outputs), readable/testable orchestration (TriggerFlow), and observable tool calls (tool_logs). It’s a better fit for teams that need reliability and maintainability after launch.

Q: Which models are supported? Is switching expensive?
A: With OpenAICompatible, you can connect OpenAI, Claude, DeepSeek, Qwen and most OpenAI‑compatible endpoints, plus local models like Llama/Qwen. The same business code can switch models without rewrites, reducing vendor lock‑in.

Q: What’s the learning curve? Where should I start?
A: The core API is straightforward—you can run your first agent in minutes. Start with Quickstart, then dive into Output Control and TriggerFlow.

Q: How do I deploy an Agently‑based service?
A: Agently doesn’t lock you into a specific deployment path. It provides async APIs and FastAPI examples. The FastAPI integration example covers SSE, WebSocket, and standard POST.

Q: Do you offer enterprise support?
A: The core framework is open‑source under Apache 2.0. For enterprise support, training, or deep collaboration, contact us via the community.

🧭 Docs Guide (Key Paths)

• Getting Started
  • Quickstart: https://agently.tech/docs/en/quickstart.html
  • Model Settings: https://agently.tech/docs/en/model-settings.html
  • Coding Agent Guide: https://agently.tech/docs/en/agent-docs.html
• Output Control (Structured Output)
  • Overview: https://agently.tech/docs/en/output-control/overview.html
  • Output Format: https://agently.tech/docs/en/output-control/format.html
  • ensure_keys: https://agently.tech/docs/en/output-control/ensure-keys.html
  • Instant Streaming: https://agently.tech/docs/en/output-control/instant-streaming.html
• Result & Streaming Events
  • Result Data: https://agently.tech/docs/en/model-response/result-data.html
  • Streaming Events: https://agently.tech/docs/en/model-response/streaming.html
• Session & Memo
  • Overview: https://agently.tech/docs/en/agent-extensions/session-memo/
  • Quickstart: https://agently.tech/docs/en/agent-extensions/session-memo/quickstart.html
• TriggerFlow Orchestration
  • Overview: https://agently.tech/docs/en/triggerflow/overview.html
  • when Branch: https://agently.tech/docs/en/triggerflow/when-branch.html
  • if / elif / else: https://agently.tech/docs/en/triggerflow/if-branch.html
  • match / case: https://agently.tech/docs/en/triggerflow/match-branch.html
  • batch: https://agently.tech/docs/en/triggerflow/batch.html
  • for_each: https://agently.tech/docs/en/triggerflow/for-each.html
  • Runtime Stream: https://agently.tech/docs/en/triggerflow/runtime-stream.html
• Tools & Extensions
  • Tools: https://agently.tech/docs/en/agent-extensions/tools.html
  • MCP: https://agently.tech/docs/en/agent-extensions/mcp.html
  • auto_func: https://agently.tech/docs/en/agent-extensions/auto-func.html
  • KeyWaiter: https://agently.tech/docs/en/agent-extensions/key-waiter.html
• Prompt Management: https://agently.tech/docs/en/prompt-management/overview.html
• Async & Settings: https://agently.tech/docs/en/async-support.html / https://agently.tech/docs/en/settings.html
• Playbook: https://agently.tech/docs/en/agent-systems/overview.html

🤝 Community

• Discussions: https://github.com/AgentEra/Agently/discussions
• Issues: https://github.com/AgentEra/Agently/issues
• WeChat Group: https://doc.weixin.qq.com/forms/AIoA8gcHAFMAScAhgZQABIlW6tV3l7QQf

📄 License

Agently is licensed under Apache 2.0.

---

Start building your production‑ready AI apps →
pip install -U agently

_{Questions? Read the docs or join the community.}