openai-agents-mantle-demo

KO | EN

gitlite — search

#javascript #hacktoberfest #react #ai #llm #go #android #machine-learning #deep-learning #docker #windows #cli #vue #awesome-list

openai-agents-mantle-demo

(★ 1)

A question-answering agent built with the OpenAI Agents SDK against Amazon Bedrock Mantle's OpenAI-compatible endpoint, developed incrementally across six versions.

openai-agents-mantle-demo 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# OpenAI Agents SDK on Amazon Bedrock Mantle

A simple question-answering agentic system built with the [OpenAI Agents SDK](https://github.com/openai/openai-agents-python) against [Amazon Bedrock Mantle](https://aws.amazon.com/bedrock/)'s OpenAI-compatible endpoint. Agents use the latest [OpenAI GPT-5.6](https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards-openai.html) model family. Agentic complexity progresses incrementally across the six script versions.

![Screengrab #1](https://raw.githubusercontent.com/garystafford/openai-agents-mantle-demo/main/screengrabs/screengrab_1.png)

![Screengrab #2](https://raw.githubusercontent.com/garystafford/openai-agents-mantle-demo/main/screengrabs/screengrab_2.png)

## What's in the repo

| Script                                                             | Adds                                                                                                 |
| ------------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------- |
| [openai-agent-demo-v1.py](openai-agent-demo-v1.py)                 | Bare `OpenAI` client pointed at the Mantle base URL                                                  |
| [openai-agent-demo-v2.py](openai-agent-demo-v2.py)                 | `Agent` + `Runner` from the Agents SDK (async)                                                       |
| [openai-agent-demo-v2-tracing.py](openai-agent-demo-v2-tracing.py) | Variant of v2 with a custom `TracingProcessor` that prints spans to the terminal (no Phoenix/Docker) |
| [openai-agent-demo-v3.py](openai-agent-demo-v3.py)                 | Agent-as-tool fact checker, Pydantic structured output, Arize Phoenix tracing                        |
| [openai-agent-demo-v4.py](openai-agent-demo-v4.py)                 | Real tools (Tavily web search, page fetch, current date), reasoning effort                           |
| [openai-agent-demo-v5.py](openai-agent-demo-v5.py)                 | Input/output guardrails, typed `RunContext`, hard search budget                                      |
| [openai-agent-demo-v6.py](openai-agent-demo-v6.py)                 | Handoffs, triage router, graceful tool-failure handling                                              |

The capabilities are cumulative: each version retains everything before it and adds one row. The staircase of checkboxes shows where each feature enters, and the final row shows where the system becomes an agent.

| Capability                              | v1  | v2  | v3  | v4  | v5  | v6  |
| --------------------------------------- | :-: | :-: | :-: | :-: | :-: | :-: |
| Agents                                  |  0  |  1  |  2  |  2  |  2  |  4  |
| Lines of code                           | 24  | 32  | 144 | 267 | 392 | 464 |
| Agents SDK framework (`Agent`/`Runner`) |     | ✅  | ✅  | ✅  | ✅  | ✅  |
| Multiple agents                         |     |     | ✅  | ✅  | ✅  | ✅  |
| Structured output (Pydantic)            |     |     | ✅  | ✅  | ✅  | ✅  |
| Streaming                               |     |     | ✅  | ✅  | ✅  | ✅  |
| Tracing (Phoenix)                       |     |     | ✅  | ✅  | ✅  | ✅  |
| Tools (web search, fetch, date)         |     |     |     | ✅  | ✅  | ✅  |
| Reasoning effort                        |     |     |     | ✅  | ✅  | ✅  |
| Guardrails (input/output)               |     |     |     |     | ✅  | ✅  |
| Typed run context + search budget       |     |     |     |     | ✅  | ✅  |
| Handoffs + triage router                |     |     |     |     |     | ✅  |
| Tool failure handling                   |     |     |     |     |     | ✅  |
| **Agent** (model-authored control flow) | ❌  | ❌  | ❌  | ✅  | ✅  | ✅  |

## Agent interaction (script v6)

```mermaid
flowchart TD
 user[User prompt] --> runner[Runner.run_streamed]
 runner --> input{Input length guardrail}
 input -- Too short --> blocked[Block request]
 input -- Valid --> triage[Triage Agent]

triage -- Factual request --> qa[Q&A Agent]
    triage -- Creative request --> creative[Creative Agent]
    creative --> creative_answer[Creative response]

qa --> date_qa[current_date tool]
 qa --> fact_check["fact_check tool (max 2 calls)"]
 fact_check --> checker[Fact Checker Agent]
 checker --> date_check[current_date tool]
 checker --> search[web_search tool]
 checker --> page[read_page tool]
 search --> tavily[Tavily API]
 page --> web[Web page]
 checker -- FactCheck --> fact_check
 fact_check -- verdict and sources --> qa
 qa --> answer[QAAnswer]

answer --> output{Output quality guardrail}
 output -- Pass --> final[Final answer]
 output -- Reject, first time --> runner
 output -- Reject again --> caveat[Return last answer with caveat]

context[(Shared RunContext 5-search budget)] -. tracks .-> search
 runner -. emits traces .-> phoenix[Phoenix]
```

## Prerequisites

- Python 3.11+ (tested against 3.13)
- An [AWS Bedrock Amazon Bedrock API key](https://docs.aws.amazon.com/bedrock/latest/userguide/api-keys-how.html)
- A [Tavily](https://tavily.com) API key (required from v4 onward)
- Docker, to run [Arize Phoenix](https://phoenix.arize.com) locally for tracing (v3+)

## Setup

Create and activate a virtual environment, then install dependencies:

```bash
python -m venv .venv
source .venv/bin/activate            # Windows: .venv\Scripts\activate
python -m pip install pip -Uq
python -m pip install -r requirements.txt #-Uq
```

## Environment variables

Create a `.env` file at the repo root:

```text
AWS_BEARER_TOKEN_BEDROCK=your-bedrock-api-key
TAVILY_API_KEY=your-tavily-key
```

The demo scripts load it automatically via `python-dotenv`. To export the values into your current shell as well:

```bash
set -a && source .env && set +a
```

## Optional: start Phoenix for tracing (v3+) in background

```bash
docker run -d -p 6006:6006 arizephoenix/phoenix:latest
```

Traces are sent to `http://localhost:6006/v1/traces`; open `http://localhost:6006` in a browser to view them.

## Run a demo

```bash
python openai-agent-demo-v1.py
# ...through v6
```

Each script is a standalone entry point — start at v1 and work up.

## Optional: render Markdown output in the terminal

The agent responses are printed in Markdown format. For a nicer terminal reading experience, you can pipe the output through [glow](https://github.com/charmbracelet/glow):

```bash
brew install glow
python openai-agent-demo-v6.py | glow -
python openai-agent-demo-v6.py | glow -w 120
```

openai-agents-mantle-demo

openai-agents-mantle-demo 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

openai-agents-mantle-demo

openai-agents-mantle-demo 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README