Metadata-Version: 2.4
Name: langchain-igpt
Version: 1.1.0
Summary: Integration package connecting iGPT and LangChain
License-Expression: MIT
License-File: LICENSE
Keywords: igpt,ai,igptai,sdk,api,llm,recall,datasources,langchain
Author: iGPT
Author-email: hello@igpt.ai
Requires-Python: >=3.10,<4.0
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3.14
Requires-Dist: aiohttp (>=3.13.3)
Requires-Dist: langchain (>=1.0.0,<2.0.0)
Requires-Dist: langchain-core (>=1.0.0,<2.0.0)
Requires-Dist: requests (>=2.32.5)
Project-URL: Documentation, https://docs.igpt.ai
Project-URL: Homepage, https://www.igpt.ai/
Project-URL: Repository, https://github.com/igptai/langchain-igpt
Description-Content-Type: text/markdown

# LangChain iGPT

LangChain integration for the **iGPT Personal Data Retrieval API** 

- Website: https://www.igpt.ai
- Documentation: https://docs.igpt.ai
- Playground: https://igpt.ai/hub/playground/

## Requirements

- Python >= 3.10

## Installation

```bash
pip install -U langchain-igpt
```

Set these environment variables (recommended):

```bash
export IGPT_API_KEY="your-api-key"
export IGPT_API_USER="your-api-user"
```

Alternatively, pass `api_key` and `user` directly to tool constructors.

### Connecting Datasources

If you receive warnings like **“No datasources found”**, it means the user has not connected any datasources yet.
Authorize and connect datasources [here](https://igpt.ai/hub/playground/connectors-authorize/).

## IgptAsk tool

Generate a response based on the input and related context. [↗](https://docs.igpt.ai/docs/api-reference/ask "API reference")

### Parameters

* `input` (string, required): The prompt/question to ask.
* `user` (string, optional if set in constructor): Unique user identifier.
* `stream` (boolean, optional, default: `false`): If `true`, returns iterable stream.
* `quality` (string, optional): Context engineering quality (default: `"cef-1-normal"`). [Read more](https://docs.igpt.ai/docs/concepts/cef).
* `output_format` (string | object, optional):

  * `"text"` (default)
  * `"json"`
  * `{ schema: <JSON Schema> }` to enforce a structured output

NOTE:
- If you set `user`, `stream`, `quality`, or `output_format` during invocation, that value will overwrite the value passed during instantiation.

### Import & Initialize

```python
from langchain_igpt import IgptAsk

tool = IgptAsk(user="user_123", api_key="ak:...", quality="cef-1-normal")
```

### Invoke directly

```python
res = tool.invoke({
    "input": "Summarize my last meeting in 5 bullet points."
})
print(res)
```

Example response:

```json
{
  "id": "...",
  "output": "...",
  "context": {
    "quality": "cef-1-normal",
    "indexed": 0.0001783,
    "datasources": [{...}]
  },
  "metadata": {
    "sources": [{...}]
  },
  "usage": {
    "input_tokens": 1272,
    "output_tokens": 281,
    "total_tokens": 1553
  }
}
```

### Invoke directly with JSON Schema

Use a schema to get consistent, machine-validated structure.

```python
output_format = {
  "strict": True,
  "schema": {
      "type": "object",
      "properties": {
        "action_items": {
          "type": "array",
          "description": "List of action items",
          "items": {
            "type": "object",
            "properties": {
              "title": { "type": "string", "description": "Short summary of the action item" },
              "owner": { "type": "string", "description": "Person responsible for the action item" },
              "due_date": { "type": "string", "format": "date", "description": "Expected completion date" }
            },
            "required": ["title", "owner", "due_date"],
            "additionalProperties": False
          }
        }
      },
      "required": ["action_items"],
      "additionalProperties": False
  }
}
res = tool.invoke({
    "input": "Extract all action items from yesterday’s board meeting.",
    "output_format": output_format
})
print(res)
```

Example response (schema):

```json
{
  "action_items": [
    {
      "title": "Approve revised Q1 budget allocation",
      "owner": "Board of Directors",
      "due_date": "2026-01-15"
    },
    {
      "title": "Approve final FY2026 strategic priorities",
      "owner": "Board of Directors",
      "due_date": "2026-01-31"
    }
  ]
}
```

### Invoke asynchronously

```python
async def main():
    result = await tool.ainvoke({"input": "List all vendor escalations from last month"})
    print(result)
asyncio.run(main())
```

### Streaming IgptAsk

For streaming responses, set `stream: True`. The tool returns an iterable that yields parsed JSON chunks.

```python
for chunk in tool.invoke({"input":"Summarize yesterday’s meeting and list action items.", "stream":True, "quality":"cef-1-normal", "output_format": "text"}):
    print(chunk, end="", flush=True)
```

### Using IgptAsk inside a LangChain Agent

This example shows how to plug the IgptAsk tool into a LangChain agent,
so the LLM can decide when to call the tool to retrieve grounded context and then
compose an answer for the user.

```python
from langchain.agents import create_agent
from langchain_openai import ChatOpenAI
from langchain_igpt import IgptAsk

ask_tool = IgptAsk()

agent = create_agent(
    model=ChatOpenAI(model="gpt-5", api_key="sk-..."),
    tools=[ask_tool],
    system_prompt=(
        "You are a helpful research assistant. Your primary job is to retrieve, "
        "summarize, and interpret user data via the igpt_ask tool."
    )
)

response = agent.invoke({
    "messages": [
        {"role": "user", "content": "Summarize key risks, decisions, and next steps from this week's meetings."}
    ]
})
print(response)
```

Streaming responses: Step-by-step updates (streams state updates for each agent step)

```python
for chunk in agent.stream(
    {
        "messages": [{"role": "user", "content": "Summarize the last 5 emails and highlight any actions I need to take."}]
    },
    stream_mode="updates"
):
    for step, data in chunk.items():
        print(f"step: {step}")
        print(f"content: {data['messages'][-1].content_blocks}")
```

Streaming responses: Token-by-token streaming

```python
for token, metadata in agent.stream(
    {"messages": [{"role": "user", "content": "Summarize PTO approvals"}]},
    stream_mode="messages",
):
    print(f"node: {metadata['langgraph_node']}")
    print(f"content: {token.content_blocks}")
    print("\n")
```

## IgptSearch tool

Search in connected datasources. [↗](https://docs.igpt.ai/docs/api-reference/search "API reference")

#### Parameters

* `query` (string, optional): Search query to execute.
* `user` (string, optional if set in constructor): Unique user identifier.
* `date_from` (string, optional): Start date filter (`YYYY-MM-DD`).
* `date_to` (string, optional): End date filter (`YYYY-MM-DD`).
* `max_results` (number, optional): Limit number of results (e.g., `50`).

NOTE:
- `query`, `date_from`, and `date_to` are intended to be provided **during invocation**
  (i.e., per-request), not as fixed constructor defaults. This lets the agent
  compute them dynamically (e.g., "last 30 days" -> concrete YYYY-MM-DD dates).
- If you set `user` or `max_results` during invocation, that value will overwrite the value passed during instantiation.

### Import & Initialize

```python
from langchain_igpt import IgptSearch

search_tool = IgptSearch(
    user="user_123",
    api_key="ak-...",
    max_results=5
)
```

### Invoke directly

```python
res = search_tool.invoke({"query": "invoices", "date_from":"2025-01-02"})
print(res)
```

Example response:

```json
{
  "id": "...",
  "context": {
    "indexed": 1,
    "datasources": [{...}]
  },
  "results": [
    {
      "id": "...",
      "timestamp": "2025-05-13T11:48:34Z",
      "type": "message",
      "from": "...",
      "content": "..."
    }
  ],
  "durations": {
    "retrieve": 47,
    "resolve": 10
  }
}
```

### Invoke asynchronously

```python
async def main():
    result = await search_tool.ainvoke({"query": "meeting notes"})
    print(result)
asyncio.run(main())
```

### Using IgptSearch inside a LangChain agent

This example shows how to use the IgptSearch tool inside a LangChain agent.

What happens when you run it:
- The agent receives the user's request: "Find invoices/receipts from the last 30 days."
- The agent uses the LLM (ChatOpenAI) to translate the request into an `IgptSearch` tool call. Typically, the LLM will:
  - set `query` to something like `"invoice OR receipt"`
  - set `date_from` to **today minus 30 days** and `date_to` to **today**, formatted as `YYYY-MM-DD`
- `IgptSearch` queries the user’s connected datasources and returns raw results + metadata.

```python
from langchain.agents import create_agent
from langchain_openai import ChatOpenAI
from langchain_igpt import IgptSearch

search_tool = IgptSearch()
agent = create_agent(
    model=ChatOpenAI(model="gpt-5", api_key="sk-..."),
    tools=[search_tool],
    system_prompt=(
        "You are a research assistant that retrieves user data using the igpt_search tool."
    )
)
response = agent.invoke({
    "messages": [
        {"role": "user", "content": "Find invoices/receipts from the last 30 days."}
    ]
})
print(response)
```

## IgptSearchRetriever

IgptSearchRetriever is a LangChain-compatible retriever that provides search over connected personal datasources via the iGPT Search API.

### Import & Initialize

```python
from langchain_igpt import IgptSearchRetriever

retriever = IgptSearchRetriever(max_results=50)
```

### Usage

```python
docs = retriever.invoke("contract renewal date", date_from="2025-01-02")
print(docs)
```

```text
[Document(metadata={'id': '<ID>', 'timestamp': '2026-01-22T14:40:12Z', 'type': 'message', 'from': '<FROM>', 'subject': 'Subscription renewal reminder Dec 17, 2025', 'source': '<SOURCE>'}, page_content='...')]
```

### Async Usage

```python
async def main():
    docs = await retriever.ainvoke("contract renewal date")
    print(docs)

asyncio.run(main())
```

## Use within a chain

```python
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_openai import ChatOpenAI
from langchain_igpt import IgptSearchRetriever

retriever = IgptSearchRetriever()

prompt = ChatPromptTemplate.from_template("""
Answer the question using only the context below.

Context:
{context}

Question:
{question}
""")

llm = ChatOpenAI(model="gpt-5", api_key="sk-...")

def format_docs(docs):
    return "\n\n".join(d.page_content for d in docs)

chain = (
    {
        "context": retriever | format_docs,
        "question": RunnablePassthrough(),
    }
    | prompt
    | llm
    | StrOutputParser()
)

res = chain.invoke("Find Invoice/Receipt")
print(res)
```

```text
('Here are all invoices/receipts found in the provided context:\n'
 '\n'
 '- Your Company (receipt)\n'
 '  - Invoice ID: ...\n'
 '  - Team: Product+Marketing\n'
 '  - Period: Jan 17, 2026 – Feb 17, 2026\n'
 '  - Total: $100.00 USD\n'
 '\n'
 '- Your Company (Invoice Receipt)\n'
 '  - Invoice Number: ...\n'
 '  - Invoice Date: 2025-11-05\n'
 '  - Customer: Jane Doe\n'
 '  - Product: Developer Membership\n'
 '  - Total: $99.00 USD (paid via credit card)\n'
 '\n'
 '- Your Business Name (WHMCS-style invoice)\n'
 '  - Invoice #: ...\n'
 '  - Date: 06/23/2018\n'
 '  - Invoiced To: John Smith\n'
 '  - Total: $37.50 USD\n'
 '  - Transaction ID: ...')
```

## License

MIT
