scriptoriumadmin/ReMe
1
0
Fork 0
mirror of https://github.com/agentscope-ai/ReMe.git synced 2026-08-28 05:25:04 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 11

update: test update_experience_pool in bfcl

Browse source
This commit is contained in:
zouying 2025-08-13 15:32:18 +08:00
parent 8cf10453fe
commit 463dc2e590
14 changed files with 1199 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
.gitignore vendored
View file

@ -30,9 +30,9 @@ cookbook/appworld/exp_result/*
file_vector_store/*
cookbook/appworld/file_vector_store/*
experiencemaker/tool/web_search_cach/*
experiencemaker/cookbook/bfcl/exp_result/*
experiencemaker/cookbook/bfcl/no_exp_result/*
experiencemaker/cookbook/bfcl/data/*
experiencemaker/cookbook/bfcl/exp_result
experiencemaker/cookbook/bfcl/no_exp_result
experiencemaker/cookbook/bfcl/data
experiencemaker/cookbook/bfcl/gorilla
experiencemaker/*.sh
experiencemaker/file_vector_store

889
experiencemaker/ExperienceMaker.egg-info/PKG-INFO Normal file
View file

@ -0,0 +1,889 @@
Metadata-Version: 2.4
Name: ExperienceMaker
Version: 0.1.1
Summary: make experience from trajectory
Author-email: experiencemaker team <experiencemaker@alibaba-inc.com>
License: Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright 2024 Alibaba Group
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Operating System :: OS Independent
Requires-Python: >=3.12
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: dashscope>=1.19.1
Requires-Dist: elasticsearch>=8.14.0
Requires-Dist: fastapi>=0.115.13
Requires-Dist: fastmcp>=2.10.6
Requires-Dist: loguru>=0.7.3
Requires-Dist: mcp>=1.9.4
Requires-Dist: numpy>=2.3.0
Requires-Dist: openai>=1.88.0
Requires-Dist: pydantic>=2.11.7
Requires-Dist: PyYAML>=6.0.2
Requires-Dist: Requests>=2.32.4
Requires-Dist: uvicorn>=0.34.3
Requires-Dist: setuptools>=75.0
Dynamic: license-file
# ExperienceMaker
<p align="center">
<img src="doc/figure/logo.jpg" alt="ExperienceMaker Logo" width="100%">
</p>
<p align="center">
<a href="https://pypi.org/project/experiencemaker/"><img src="https://img.shields.io/badge/python-3.12+-blue" alt="Python Version"></a>
<a href="https://pypi.org/project/experiencemaker/"><img src="https://img.shields.io/badge/pypi-v0.1.1-blue?logo=pypi" alt="PyPI Version"></a>
<a href="./LICENSE"><img src="https://img.shields.io/badge/license-Apache--2.0-black" alt="License"></a>
<a href="https://github.com/modelscope/ExperienceMaker"><img src="https://img.shields.io/github/stars/modelscope/ExperienceMaker?style=social" alt="GitHub Stars"></a>
</p>
<p align="center">
<strong>A comprehensive framework to make & reuse & share experience for AI agent</strong><br>
<em>Empowering agents to learn from the past and excel in the future</em>
</p>
---
## 📰 What's New
- **[2025-08]** 🚀 MCP is now available! → [Quick Start Guide](./doc/mcp_quick_start.md)
- **[2025-07]** 🎉 ExperienceMaker v0.1.1 is now available on [PyPI](https://pypi.org/project/experiencemaker/)!
- **[2025-07]** 📚 Complete documentation and quick start guides released
- **[2025-06]** 🚀 Multi-backend vector store support (Elasticsearch & ChromaDB)
---
## 🚀 What's Next
- **Pre-built Experience Libraries**: Domain repositories (Finance/Coding/Education/Research) + community marketplace
- **Rich Experience Formats**: Executable code/tool configs/pipeline templates/workflows
- **Experience Validation**: Quality analysis + cross-task effectiveness + auto-refinement
- **Universal Trajectory Extraction**: Raw logs/multimodal data/execution traces → experiences
---
## 🌟 What is ExperienceMaker?
ExperienceMaker is a framework that transforms how AI agents learn and improve through **experience-driven intelligence**.
By automatically extracting, storing, and intelligently reusing experiences from agent trajectories, it enables continuous learning and progressive skill enhancement.
### ✨ Core Capabilities
#### 🔍 **Intelligent Experience Summarizer**
- **Success Pattern Recognition**: Identify what works and understand the underlying principles
- **Failure Analysis**: Learn from mistakes to avoid repeating them in future tasks
- **Comparative Insights**: Understand the critical differences between successful and failed approaches
- **Multistep Trajectory Processing**: Break down complex tasks into learnable, actionable segments
#### 🎯 **Smart Experience Retriever**
- **Semantic Search**: Find relevant experiences using advanced embedding models and semantic understanding
- **Context-Aware Ranking**: Prioritize the most applicable experiences for current task contexts
- **Dynamic Rewriting**: Intelligently adapt experiences to fit new situations and requirements
- **Multi-modal Support**: Handle various input types including query, messages
#### 🗄️ **Scalable Experience Management**
- **Multiple Storage Backends**: Choose from Elasticsearch (production-ready), ChromaDB (development), or file-based storage (testing)
- **Workspace Isolation**: Organize experiences by projects, domains, or teams with complete separation
- **Deduplication & Validation**: Ensure high-quality, unique experience storage with automated quality control
- **Batch Operations**: Efficiently handle large-scale experience processing with optimized performance
#### 🔧 **Developer-Friendly Architecture**
- **REST API Interface**: Seamless integration with existing systems through clean API design
- **Modular Pipeline Design**: Compose custom workflows from atomic operations with maximum flexibility
- **Flexible Configuration**: YAML files and command-line overrides for easy customization
- **Experience Store**: Ready-to-use out of the box — there’s no need for you to manually summarize experiences. You can directly leverage existing, comprehensive experience datasets to greatly enhance your agent’s capabilities.
<p align="center">
<img src="doc/figure/framework.png" alt="ExperienceMaker Architecture" width="70%">
</p>
---
## 🛠️ Installation
### Option 1: Install from PyPI (Recommended)
```bash
pip install experiencemaker
```
### Option 2: Install from Source
```bash
git clone https://github.com/modelscope/ExperienceMaker.git
cd ExperienceMaker
pip install .
```
## ⚙️ Environment Setup
Create a `.env` file in your project root directory:
```bash
# Required: LLM API configuration
LLM_API_KEY="sk-xxx"
LLM_BASE_URL="https://xxx.com/v1"
# Required: Embedding model configuration
EMBEDDING_MODEL_API_KEY="sk-xxx"
EMBEDDING_MODEL_BASE_URL="https://xxx.com/v1"
# Optional: Elasticsearch configuration (if using Elasticsearch backend)
```
## 🚀 Quick Start
### 🌐 HTTP Service
For testing and development, use the `local_file` backend:
```bash
experiencemaker \
http_service.port=8001 \
llm.default.model_name=qwen3-32b \
embedding_model.default.model_name=text-embedding-v4 \
vector_store.default.backend=local_file
```
💡 **Pro Tip**: Check out our [Configuration Guide](./doc/configuration_guide.md) for detailed configuration topics
including custom pipelines, operation parameters, and advanced configuration methods.
The service will start on `http://localhost:8001`
### 🔌 MCP Server
ExperienceMaker now supports Model Context Protocol (MCP) for seamless integration with MCP-compatible clients like Claude Desktop:
```bash
experiencemaker_mcp \
mcp_transport=stdio \
llm.default.model_name=qwen3-32b \
embedding_model.default.model_name=text-embedding-v4 \
vector_store.default.backend=local_file
```
For SSE transport (Server-Sent Events):
```bash
experiencemaker_mcp \
mcp_transport=sse \
http_service.port=8001 \
llm.default.model_name=qwen3-32b \
embedding_model.default.model_name=text-embedding-v4 \
vector_store.default.backend=local_file
```
🔗 **For detailed MCP setup and usage examples**, see our [MCP Quick Start Guide](./doc/mcp_quick_start.md).
### 🔍 Production Setup with Elasticsearch Backend
```bash
experiencemaker \
http_service.port=8001 \
llm.default.model_name=qwen3-32b \
embedding_model.default.model_name=text-embedding-v4 \
vector_store.default.backend=elasticsearch
```
**Setup Elasticsearch:**
```bash
export ES_HOSTS="http://localhost:9200"
# Quick setup using Elastic's official script
curl -fsSL https://elastic.co/start-local | sh
```
📖 **Need Help?** Refer to [Vector Store Setup](./doc/vector_store_setup.md) for comprehensive deployment guidance.
## 📝 Your First ExperienceMaker Script
Here's how to get started!
Note the `workspace_id` serves as your experience storage namespace. Experiences in different workspaces remain completely isolated and cannot access each other.
### 📊 Call Summarizer Examples
Transform conversation trajectories into valuable experiences using batch summarization. Each trajectory contains:
- **Message**: Complete conversation history between user and agent
- **Score**: Performance rating (0-1 scale, where 0=failure, 1=success)
The summarizer analyzes these trajectories to extract actionable insights and patterns for future interactions.
<details open>
<summary><b>Python</b></summary>
```python
import requests
response = requests.post(url="http://0.0.0.0:8001/summarizer", json={
"workspace_id": "test_workspace",
"traj_list": [
{"messages": [{"role": "user", "content": "hello world"}], "score": 1.0}
]
})
experience_list = response.json()["experience_list"]
for experience in experience_list:
print(experience)
```
</details>
<details>
<summary><b>curl</b></summary>
```bash
curl -X POST "http://0.0.0.0:8001/summarizer" \
-H "Content-Type: application/json" \
-d '{
"workspace_id": "test_workspace",
"traj_list": [
{
"messages": [{"role": "user", "content": "hello world"}],
"score": 1.0
}
]
}'
```
</details>
<details>
<summary><b>Node.js</b></summary>
```javascript
const fetch = require('node-fetch');
// or: import fetch from 'node-fetch';
async function callSummarizer() {
try {
const response = await fetch('http://0.0.0.0:8001/summarizer', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
body: JSON.stringify({
workspace_id: "test_workspace",
traj_list: [
{
messages: [{ role: "user", content: "hello world" }],
score: 1.0
}
]
})
});
const data = await response.json();
const experienceList = data.experience_list;
experienceList.forEach(experience => {
console.log(experience);
});
} catch (error) {
console.error('Error:', error);
}
}
callSummarizer();
```
</details>
### 🔍 Call Retriever Examples
Intelligently search and retrieve the most relevant experiences from your workspace to enhance decision-making. The retriever:
- **Finds** the top-k most similar experiences based on semantic similarity to your query
- **Returns** pre-assembled context ready for immediate use, or raw experience data for custom processing
- **Leverages** your workspace's accumulated knowledge to provide contextually relevant insights
<details open>
<summary><b>Python</b></summary>
```python
import requests
response = requests.post(url="http://0.0.0.0:8001/retriever", json={
"workspace_id": "test_workspace",
"query": "what is the meaning of life?",
"top_k": 1,
})
experience_merged: str = response.json()["experience_merged"]
print(f"experience_merged={experience_merged}")
```
</details>
<details>
<summary><b>curl</b></summary>
```bash
curl -X POST "http://0.0.0.0:8001/retriever" \
-H "Content-Type: application/json" \
-d '{
"workspace_id": "test_workspace",
"query": "what is the meaning of life?",
"top_k": 1
}'
```
</details>
<details>
<summary><b>Node.js</b></summary>
```javascript
const fetch = require('node-fetch');
// or: import fetch from 'node-fetch';
async function callRetriever() {
try {
const response = await fetch('http://0.0.0.0:8001/retriever', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
body: JSON.stringify({
workspace_id: "test_workspace",
query: "what is the meaning of life?",
top_k: 1
})
});
const data = await response.json();
const experienceMerged = data.experience_merged;
console.log(`experience_merged=${experienceMerged}`);
} catch (error) {
console.error('Error:', error);
}
}
callRetriever();
```
</details>
### 💾 Dump Experiences From Vector Store
Export and backup your valuable experience data for archival, analysis, or migration purposes. This operation:
- **Extracts** all experiences from the specified workspace in the vector store
- **Saves** them to a structured JSONL file at `{path}/{workspace_id}.jsonl`
- **Preserves** complete experience metadata and embeddings for future restoration
<details open>
<summary><b>Python</b></summary>
```python
import requests
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
"workspace_id": "test_workspace",
"action": "dump",
"path": "./",
})
print(response.json())
```
</details>
<details>
<summary><b>curl</b></summary>
```bash
curl -X POST "http://0.0.0.0:8001/vector_store" \
-H "Content-Type: application/json" \
-d '{
"workspace_id": "test_workspace",
"action": "dump",
"path": "./"
}'
```
</details>
<details>
<summary><b>Node.js</b></summary>
```javascript
const fetch = require('node-fetch');
// or: import fetch from 'node-fetch';
async function dumpExperiences() {
try {
const response = await fetch('http://0.0.0.0:8001/vector_store', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
body: JSON.stringify({
workspace_id: "test_workspace",
action: "dump",
path: "./"
})
});
const data = await response.json();
console.log(data);
} catch (error) {
console.error('Error:', error);
}
}
dumpExperiences();
```
</details>
### 📥 Load Experiences To Vector Store
Import and restore previously exported experience data to populate your workspace with existing knowledge. This operation:
- **Reads** experience data from the JSONL file located at `{path}/{workspace_id}.jsonl`
- **Reconstructs** the vector embeddings and indexes them in the specified workspace
- **Enables** immediate access to imported experiences for retrieval and decision-making
<details open>
<summary><b>Python</b></summary>
```python
import requests
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
"workspace_id": "test_workspace",
"action": "load",
"path": "./",
})
print(response.json())
```
</details>
<details>
<summary><b>curl</b></summary>
```bash
curl -X POST "http://0.0.0.0:8001/vector_store" \
-H "Content-Type: application/json" \
-d '{
"workspace_id": "test_workspace",
"action": "load",
"path": "./"
}'
```
</details>
<details>
<summary><b>Node.js</b></summary>
```javascript
const fetch = require('node-fetch');
// or: import fetch from 'node-fetch';
async function loadExperiences() {
try {
const response = await fetch('http://0.0.0.0:8001/vector_store', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
body: JSON.stringify({
workspace_id: "test_workspace",
action: "load",
path: "./"
})
});
const data = await response.json();
console.log(data);
} catch (error) {
console.error('Error:', error);
}
}
loadExperiences();
```
</details>
💡 **Need More Advanced Operations?** For additional workspace management features(e.g. delete_workspace,
copy_workspace), advanced configuration options, and troubleshooting guidance, check out our
comprehensive [Quick Start Guide](./cookbook/simple_demo/quick_start.md).
🎭 **Want to See It in Action?** We've prepared a [simple react agent](./cookbook/simple_demo/simple_demo.py) that demonstrates how to enhance agent capabilities by integrating summarizer and retriever components, achieving significantly better performance.
---
## 🧪 Experiments
### 🌍 Experiment on Appworld
We test ExperienceMaker on Appworld with qwen3-8b:
| Method | pass@1 | pass@2 | pass@4 |
|--------------------------------|-----------|-------------|-----------|
| w/o ExperienceMaker (baseline) | 0.083 | 0.140 | 0.228 |
| **w ExperienceMaker** | | | |
|experience(Direct Use) | **0.109** | **0.175** | **0.281** |
Pass@K measures the probability that at least one out of K generated samples successfully completes the task (achieves score=1).
The current experiments use an internal AppWorld environment which may have slight discrepancies, and we will soon update with experimental results from the standard AppWorld environment.
You may find more details to reproduce this experiment in [quickstart.md](cookbook/appworld/quickstart.md)
### 🧊 Experiment on Frozenlake
| without experience | with experience |
|:-------------------------------------------------------------------------------------------:|:-------------------------------------------:|
| <p align="center"><img src="doc/figure/frozenlake_failure.gif" alt="GIF 1" width="30%"></p> | <p align="center"><img src="doc/figure/frozenlake_success.gif" alt="GIF 2" width="30%"></p>
We test on 100 random frozenlake map with qwen3-8b:
| Method | pass rate |
|-------------------------------|------------------|
| w/o ExperienceMaker (baseline) | 0.66 |
| **w ExperienceMaker** | |
| [1] experience(Direct Use) | 0.72 **(+9.1%)** |
| [2] experience(LLM Rewritten) | 0.72 **(+9.1%)** |
We also noticed that in such simple scenarios, not using LLM rewriting may actually yield better results.
Therefore, in some simple scenarios, you can also try disabling LLM rewriting by simply changing the following in default_config.yaml:
```yaml
rewrite_experience_op:
params:
enable_llm_rewrite: false # change this to false
```
You may find more details to reproduce this experiment in [quickstart.md](cookbook/frozenlake/quickstart.md)
### 🔧 Experiment on BFCL-V3
Coming Soon! Stay tuned for comprehensive evaluation results.
---
## 🏪 Ready-made Experience Store
ExperienceMaker provides pre-built experience libraries to jumpstart your agent's capabilities.
You can directly load these curated experiences into your workspace and start benefiting from accumulated knowledge
immediately.
### 📦 Available Experience Libraries
- **`appworld_v1.jsonl`**: Comprehensive experiences from Appworld agent interactions, covering complex task planning
and execution patterns
- **`bfcl_v1.jsonl`**: Function calling experiences from Berkeley Function-Calling Leaderboard tasks
### 🚀 Quick Start with Pre-built Experiences
Here's how to load and use the Appworld experience library:
#### Step 1: Load Pre-built Experiences
<details open>
<summary><b>Python</b></summary>
```python
import requests
# Load Appworld experiences into your workspace
response = requests.post(url="http://0.0.0.0:8001/vector_store", json={
"workspace_id": "appworld_v1",
"action": "load",
"path": "./library/",
})
print(f"loading result result={response.json()}")
```
</details>
<details>
<summary><b>curl</b></summary>
```bash
curl -X POST "http://0.0.0.0:8001/vector_store" \
-H "Content-Type: application/json" \
-d '{
"workspace_id": "appworld_v1",
"action": "load",
"path": "./library/"
}'
```
</details>
#### Step 2: Retrieve Relevant Experiences
Now you can query the loaded experiences to get contextual guidance for your tasks:
<details open>
<summary><b>Python</b></summary>
```python
import requests
# Query for app interaction experiences
response = requests.post(url="http://0.0.0.0:8001/retriever", json={
"workspace_id": "appworld_v1",
"query": "How to navigate to settings and update user profile information?",
"top_k": 1,
})
experience_merged = response.json()["experience_merged"]
print(f"Retrieved experiences: {experience_merged}")
```
</details>
<details>
<summary><b>curl</b></summary>
```bash
curl -X POST "http://0.0.0.0:8001/retriever" \
-H "Content-Type: application/json" \
-d '{
"workspace_id": "appworld_v1",
"query": "How to navigate to settings and update user profile information?",
"top_k": 1
}'
```
</details>
---
## 📚 Additional Resources
- **[Quick Start](./cookbook/simple_demo/quick_start.md)**: This guide will help you get started with ExperienceMaker quickly using practical examples.
- **[Vector Store Setup](./doc/vector_store_setup.md)**: Complete production deployment guide
- **[Configuration Guide](./doc/configuration_guide.md)**: Describes all available command-line parameters for ExperienceMaker Service
- **[Operations Documentation](./doc/operations_documentation.md)**: Comprehensive operations configuration reference
- **[Example Collection](./cookbook)**: Practical examples and use cases
- **[Future RoadMap](./doc/future_roadmap.md)**: Our vision and upcoming features
---
## 🤝 Contributing
We warmly welcome contributions from the community! Here's how you can help make ExperienceMaker even better:
### 🐛 **Report Issues**
- Bug reports with detailed reproduction steps
- Feature requests and enhancement suggestions
- Documentation improvements and clarifications
- Performance optimization ideas
### 💻 **Code Contributions**
- New operations and tools development
- Backend implementations and optimizations
- API enhancements and new endpoints
- Test coverage improvements and quality assurance
### 📝 **Documentation**
- Usage examples and comprehensive tutorials
- Best practices guides and design patterns
- Translation and localization efforts
---
## 📄 Citation
If you use ExperienceMaker in your research or projects, please cite:
```bibtex
@software{ExperienceMaker,
title = {ExperienceMaker: A Comprehensive Framework for AI Agent Experience Generation and Reuse},
author = {The ExperienceMaker Team},
url = {https://github.com/modelscope/ExperienceMaker},
month = {08},
year = {2025},
}
```
---
## ⚖️ License
This project is licensed under the Apache License 2.0 - see the [LICENSE](./LICENSE) file for details.
---

103
experiencemaker/ExperienceMaker.egg-info/SOURCES.txt Normal file
View file

@ -0,0 +1,103 @@
LICENSE
README.md
pyproject.toml
ExperienceMaker.egg-info/PKG-INFO
ExperienceMaker.egg-info/SOURCES.txt
ExperienceMaker.egg-info/dependency_links.txt
ExperienceMaker.egg-info/entry_points.txt
ExperienceMaker.egg-info/requires.txt
ExperienceMaker.egg-info/top_level.txt
experiencemaker/__init__.py
experiencemaker/app.py
experiencemaker/mcp_server.py
experiencemaker/config/__init__.py
experiencemaker/config/config_parser.py
experiencemaker/config/default_config.yaml
experiencemaker/config/mock_config.yaml
experiencemaker/config/simple_config.yaml
experiencemaker/embedding_model/__init__.py
experiencemaker/embedding_model/base_embedding_model.py
experiencemaker/embedding_model/openai_compatible_embedding_model.py
experiencemaker/enumeration/__init__.py
experiencemaker/enumeration/agent_state.py
experiencemaker/enumeration/chunk_enum.py
experiencemaker/enumeration/http_enum.py
experiencemaker/enumeration/role.py
experiencemaker/llm/__init__.py
experiencemaker/llm/base_llm.py
experiencemaker/llm/openai_compatible_llm.py
experiencemaker/op/__init__.py
experiencemaker/op/base_op.py
experiencemaker/op/mock_op.py
experiencemaker/op/prompt_mixin.py
experiencemaker/op/react/__init__.py
experiencemaker/op/react/react_v1_op.py
experiencemaker/op/react/react_v1_prompt.yaml
experiencemaker/op/retriever/__init__.py
experiencemaker/op/retriever/build_query_op.py
experiencemaker/op/retriever/build_query_prompt.yaml
experiencemaker/op/retriever/merge_experience_op.py
experiencemaker/op/retriever/rerank_experience_op.py
experiencemaker/op/retriever/rerank_experience_prompt.yaml
experiencemaker/op/retriever/rewrite_experience_op.py
experiencemaker/op/retriever/rewrite_experience_prompt.yaml
experiencemaker/op/summarizer/__init__.py
experiencemaker/op/summarizer/comparative_extraction_op.py
experiencemaker/op/summarizer/comparative_extraction_prompt.yaml
experiencemaker/op/summarizer/experience_deduplication_op.py
experiencemaker/op/summarizer/experience_validation_op.py
experiencemaker/op/summarizer/experience_validation_prompt.yaml
experiencemaker/op/summarizer/failure_extraction_op.py
experiencemaker/op/summarizer/failure_extraction_prompt.yaml
experiencemaker/op/summarizer/pdf_preprocess_op.py
experiencemaker/op/summarizer/simple_comparative_summary_op.py
experiencemaker/op/summarizer/simple_comparative_summary_prompt.yaml
experiencemaker/op/summarizer/simple_summary_op.py
experiencemaker/op/summarizer/simple_summary_prompt.yaml
experiencemaker/op/summarizer/success_extraction_op.py
experiencemaker/op/summarizer/success_extraction_prompt.yaml
experiencemaker/op/summarizer/trajectory_preprocess_op.py
experiencemaker/op/summarizer/trajectory_segmentation_op.py
experiencemaker/op/summarizer/trajectory_segmentation_prompt.yaml
experiencemaker/op/vector_store/__init__.py
experiencemaker/op/vector_store/recall_vector_store_op.py
experiencemaker/op/vector_store/update_vector_store_op.py
experiencemaker/op/vector_store/vector_store_action_op.py
experiencemaker/pipeline/__init__.py
experiencemaker/pipeline/pipeline.py
experiencemaker/pipeline/pipeline_context.py
experiencemaker/schema/__init__.py
experiencemaker/schema/app_config.py
experiencemaker/schema/experience.py
experiencemaker/schema/message.py
experiencemaker/schema/request.py
experiencemaker/schema/response.py
experiencemaker/schema/vector_node.py
experiencemaker/service/__init__.py
experiencemaker/service/experience_maker_client.py
experiencemaker/service/experience_maker_service.py
experiencemaker/service/mcp_client.py
experiencemaker/tool/__init__.py
experiencemaker/tool/base_tool.py
experiencemaker/tool/code_tool.py
experiencemaker/tool/dashscope_search_tool.py
experiencemaker/tool/mcp_tool.py
experiencemaker/tool/tavily_search_tool.py
experiencemaker/tool/terminate_tool.py
experiencemaker/utils/__init__.py
experiencemaker/utils/common_utils.py
experiencemaker/utils/file_handler.py
experiencemaker/utils/http_client.py
experiencemaker/utils/op_utils.py
experiencemaker/utils/registry.py
experiencemaker/utils/singleton.py
experiencemaker/utils/timer.py
experiencemaker/vector_store/__init__.py
experiencemaker/vector_store/base_vector_store.py
experiencemaker/vector_store/chroma_vector_store.py
experiencemaker/vector_store/es_vector_store.py
experiencemaker/vector_store/file_vector_store.py
test/test1.py
test/test2.py
test/test3.py
test/test4.py

1
experiencemaker/ExperienceMaker.egg-info/dependency_links.txt Normal file
View file

@ -0,0 +1 @@

3
experiencemaker/ExperienceMaker.egg-info/entry_points.txt Normal file
View file

@ -0,0 +1,3 @@
[console_scripts]
experiencemaker = experiencemaker.app:main
experiencemaker_mcp = experiencemaker.mcp_server:main

13
experiencemaker/ExperienceMaker.egg-info/requires.txt Normal file
View file

@ -0,0 +1,13 @@
dashscope>=1.19.1
elasticsearch>=8.14.0
fastapi>=0.115.13
fastmcp>=2.10.6
loguru>=0.7.3
mcp>=1.9.4
numpy>=2.3.0
openai>=1.88.0
pydantic>=2.11.7
PyYAML>=6.0.2
Requests>=2.32.4
uvicorn>=0.34.3
setuptools>=75.0

1
experiencemaker/ExperienceMaker.egg-info/top_level.txt Normal file
View file

@ -0,0 +1 @@
experiencemaker

15
experiencemaker/cookbook/bfcl/bfcl_agent.py
View file

@ -63,6 +63,7 @@ class BFCLAgent:
num_runs: int = 1,
enable_thinking: bool = False,
use_experience: bool = False,
use_fixed_experience: bool = True,
experience_base_url: str = "http://0.0.0.0:8001/",
experience_workspace_id: str = "bfcl_8b_0725"):
@ -79,6 +80,7 @@ class BFCLAgent:
self.num_runs: int = num_runs
self.enable_thinking: bool = enable_thinking
self.use_experience: bool = use_experience
self.use_fixed_experience: bool = use_fixed_experience
self.experience_base_url: str = experience_base_url
self.experience_workspace_id: str = experience_workspace_id
@ -135,11 +137,11 @@ class BFCLAgent:
def update_experience(self, trajectories):
response = requests.post(url=self.experience_base_url + "summarizer", json={
"workspace_id": self.experience_workspace_id,
"trajectories": trajectories,
"traj_list": trajectories,
})
response.raise_for_status()
response = response.json()
return response["experiences"]
print(f"add new experiences: {response["experience_list"]}")
def call_llm(self, messages: list, tool_schemas: list[dict]) -> str:
for i in range(100):
@ -517,9 +519,12 @@ class BFCLAgent:
break
reward = self.get_reward(run_id, task_index)
# if reward == 1:
#
# self.update_experience([process_msg_to_trajectory(task_id, msg, reward)]) # selectively add experiences when succeed
if reward == 1 and not self.use_fixed_experience:
self.update_experience([{
"task_id":task_id,
"messages":self.history[run_id][task_index],
"score":reward
}]) # selectively add experiences when succeed
t_result = {
"run_id": run_id,

9
experiencemaker/cookbook/bfcl/run_bfcl.py
View file

@ -22,6 +22,7 @@ def run_agent(dataset_name: str,
data_path: str = "data/multiturn_data_base_val.jsonl",
answer_path: Path = Path("data/possible_answer"),
use_experience: bool = False,
use_fixed_experience: bool = True,
enable_thinking: bool = False,
experience_base_url: str = "http://0.0.0.0:8001/",
experience_workspace_id: str = "bfcl_8b_0725"):
@ -51,6 +52,7 @@ def run_agent(dataset_name: str,
model_name=model_name,
num_runs=num_runs,
use_experience=use_experience,
use_fixed_experience=use_fixed_experience,
enable_thinking=enable_thinking,
experience_base_url=experience_base_url,
experience_workspace_id=experience_workspace_id
@ -82,6 +84,7 @@ def run_agent(dataset_name: str,
answer_path=answer_path,
enable_thinking=enable_thinking,
use_experience=use_experience,
use_fixed_experience=use_fixed_experience,
experience_base_url=experience_base_url,
experience_workspace_id=experience_workspace_id)
task_results = agent.execute()
@ -95,6 +98,7 @@ def main():
max_workers = 4
num_runs = 4 # Run each task 4 times
use_experience = True
use_fixed_experience = True
experience_base_url = "http://0.0.0.0:8001/"
experience_workspace_id = "bfcl_v1"
if max_workers > 1:
@ -102,14 +106,15 @@ def main():
for run_id in range(num_runs):
run_agent(
dataset_name="bfcl-multi-turn-base-val",
experiment_suffix=f"0812-w-exp-extract-compare-recall",
experiment_suffix=f"0813-w-exp-w-think-update-test",
model_name="qwen3-8b",
max_workers=max_workers,
num_runs=1,
data_path="data/multiturn_data_base_val.jsonl",
answer_path=Path("data/possible_answer"),
enable_thinking=False,
enable_thinking=True,
use_experience=use_experience,
use_fixed_experience=use_fixed_experience,
experience_base_url=experience_base_url,
experience_workspace_id=experience_workspace_id,
)

2
experiencemaker/cookbook/bfcl/run_exp_statistic.py
View file

@ -61,7 +61,7 @@ def get_possible_k_values(total_runs: int) -> list:
def run_exp_statistic():
path: Path = Path(f"./no_exp_result/qwen3-8b")
path: Path = Path(f"./exp_result/qwen3-8b")
# Store results for all experiments
all_results = {}

0
experiencemaker/experiencemaker/op/manager/__init__.py Normal file
View file

162
experiencemaker/experiencemaker/op/manager/experience_deletion_op.py Normal file
View file

@ -0,0 +1,162 @@
from typing import List
from loguru import logger
from experiencemaker.op import OP_REGISTRY
from experiencemaker.op.base_op import BaseOp
from experiencemaker.schema.experience import BaseExperience
from experiencemaker.schema.request import ManagerRequest
@OP_REGISTRY.register()
class ExperienceDeletionOp(BaseOp):
current_path: str = __file__
def execute(self):
"""Remove low-utility experiences"""
request: ManagerRequest = self.context.request
experiences: List[BaseExperience] = self.context.response.experience_list
if not experiences:
logger.info("No experiences found for deduplication")
return
logger.info(f"Starting deduplication for {len(experiences)} experiences")
# Perform deduplication
deduplicated_experiences = self._deduplicate_experiences(experiences)
logger.info(f"Deduplication complete: {len(deduplicated_experiences)} deduplicated experiences out of {len(experiences)}")
# Update context
self.context.response.experience_list = deduplicated_experiences
def _deduplicate_experiences(self, experiences: List[BaseExperience]) -> List[BaseExperience]:
"""Remove duplicate experiences"""
if not experiences:
return experiences
similarity_threshold = self.op_params.get("similarity_threshold", 0.5)
workspace_id = self.context.request.workspace_id if hasattr(self.context, 'request') else None
unique_experiences = []
# Get existing experience embeddings
existing_embeddings = self._get_existing_experience_embeddings(workspace_id)
for experience in experiences:
# Generate embedding for current experience
current_embedding = self._get_experience_embedding(experience)
if current_embedding is None:
logger.warning(f"Failed to generate embedding for experience: {str(experience.when_to_use)[:50]}...")
continue
# Check similarity with existing experiences
if self._is_similar_to_existing_experiences(current_embedding, existing_embeddings, similarity_threshold):
logger.debug(f"Skipping similar experience: {str(experience.when_to_use)[:50]}...")
continue
# Check similarity with current batch experiences
if self._is_similar_to_current_experiences(current_embedding, unique_experiences, similarity_threshold):
logger.debug(f"Skipping duplicate in current batch: {str(experience.when_to_use)[:50]}...")
continue
# Add to unique experiences list
unique_experiences.append(experience)
logger.debug(f"Added unique experience: {str(experience.when_to_use)[:50]}...")
return unique_experiences
def _get_existing_experience_embeddings(self, workspace_id: str) -> List[List[float]]:
"""Get embeddings of existing experiences"""
try:
if not hasattr(self, 'vector_store') or not self.vector_store or not workspace_id:
return []
# Query existing experience nodes
existing_nodes = self.vector_store.search(
query="...", # Empty query to get all
workspace_id=workspace_id,
top_k=self.op_params.get("max_existing_experiences", 1000)
)
# Extract embeddings
existing_embeddings = []
for node in existing_nodes:
if hasattr(node, 'embedding') and node.embedding:
existing_embeddings.append(node.embedding)
logger.debug(f"Retrieved {len(existing_embeddings)} existing experience embeddings from workspace {workspace_id}")
return existing_embeddings
except Exception as e:
logger.warning(f"Failed to retrieve existing experience embeddings: {e}")
return []
def _get_experience_embedding(self, experience: BaseExperience) -> List[float]:
"""Generate embedding for experience"""
try:
if not hasattr(self, 'vector_store') or not self.vector_store:
return None
# Combine experience description and content for embedding
text_for_embedding = f"{experience.when_to_use} {experience.content}"
embeddings = self.vector_store.embedding_model.get_embeddings([text_for_embedding])
if embeddings and len(embeddings) > 0:
return embeddings[0]
else:
logger.warning("Empty embedding generated for experience")
return None
except Exception as e:
logger.error(f"Error generating embedding for experience: {e}")
return None
def _is_similar_to_existing_experiences(self, current_embedding: List[float],
existing_embeddings: List[List[float]],
threshold: float) -> bool:
"""Check if current embedding is similar to existing embeddings"""
for existing_embedding in existing_embeddings:
similarity = self._calculate_cosine_similarity(current_embedding, existing_embedding)
if similarity > threshold:
logger.debug(f"Found similar existing experience with similarity: {similarity:.3f}")
return True
return False
def _is_similar_to_current_experiences(self, current_embedding: List[float],
current_experiences: List[BaseExperience],
threshold: float) -> bool:
for existing_experience in current_experiences:
existing_embedding = self._get_experience_embedding(existing_experience)
if existing_embedding is None:
continue
similarity = self._calculate_cosine_similarity(current_embedding, existing_embedding)
if similarity > threshold:
logger.debug(f"Found similar experience in current batch with similarity: {similarity:.3f}")
return True
return False
def _calculate_cosine_similarity(self, embedding1: List[float], embedding2: List[float]) -> float:
"""Calculate cosine similarity"""
try:
import numpy as np
vec1 = np.array(embedding1)
vec2 = np.array(embedding2)
# Calculate cosine similarity
dot_product = np.dot(vec1, vec2)
norm1 = np.linalg.norm(vec1)
norm2 = np.linalg.norm(vec2)
if norm1 == 0 or norm2 == 0:
return 0.0
return dot_product / (norm1 * norm2)
except Exception as e:
logger.error(f"Error calculating cosine similarity: {e}")
return 0.0

2
experiencemaker/experiencemaker/op/vector_store/recall_vector_store_op.py
View file

@ -25,7 +25,7 @@ class RecallVectorStoreOp(BaseOp):
nodes: List[VectorNode] = self.vector_store.search(query=query,
workspace_id=request.workspace_id,
top_k=request.top_k)
# convert to experience, filter duplicate
experience_list: List[BaseExperience] = []
experience_content_list: List[str] = []

5
experiencemaker/experiencemaker/schema/request.py
View file

@ -20,6 +20,11 @@ class SummarizerRequest(BaseRequest):
traj_list: List[Trajectory] = Field(default_factory=list)
class ManagerRequest(BaseRequest):
freq_threshold: int = Field(default=10)
utility_threshold: float = Field(default=0.6)
class VectorStoreRequest(BaseRequest):
action: str = Field(default="")
src_workspace_id: str = Field(default="")