Prompt Optimization & Context Window Management

Large OpenAPI specifications (e.g. Crossref, GitHub, Slack) contain extensive operation descriptions, parameter documentation, and schema definitions. Passing raw specifications or unoptimized tools directly into an LLM context window can lead to context exhaustion, high latency, and elevated API costs.

langchain_openapi includes built-in prompt optimization tools to compress tool descriptions and filter unnecessary operations before tool creation.

---

The OpenAPIToolkitConfig Configuration Object

Customize tool creation behavior by initializing an OpenAPIToolkitConfig object and passing it to OpenAPIToolkit:

from langchain_openapi_tools import OpenAPIToolkit, OpenAPIToolkitConfig

config = OpenAPIToolkitConfig(
    description_mode="compact",
    compress_descriptions=True,
    include_tags=["Works"],
    tool_description_overrides={
        "get_works": "Search scholarly publications in Crossref."
    },
)

toolkit = OpenAPIToolkit.from_url(
    "https://api.crossref.org/swagger-docs", config=config
)

Alternatively, pass keyword arguments directly for convenience:

toolkit = OpenAPIToolkit.from_url(
    "https://api.crossref.org/swagger-docs",
    description_mode="compact",
    compress_descriptions=True,
    include_tags=["Works"],
)

---

Tool Description Modes (description_mode)

langchain_openapi supports three description modes:

1. full (Default)

Includes the full operation summary, full description, HTTP method, and path:

Search scholarly works registered with Crossref.

Returns metadata for matching DOI entries.

HTTP Method: GET
Path: /works

2. compact

Includes only a single-line summary and a concise parameter list, omitting HTTP method, path, response schemas, and examples:

Search scholarly works.
Parameters: query (query, required), rows (query), offset (query)

3. minimal

Includes a single sentence. Best for APIs with dozens of endpoints to minimize context usage:

Search scholarly works.

---

Description Compression (compress_descriptions=True)

Enable description compression to automatically: - Remove duplicate sentences (e.g., when summary repeats description). - Strip redundant formatting, empty sections, and extra whitespace. - Omit response schemas and non-essential examples.

toolkit = OpenAPIToolkit.from_url(url, compress_descriptions=True)

---

Description Overrides & Callback Builders

Tool Description Overrides

Directly override tool descriptions for specific operations:

toolkit = OpenAPIToolkit.from_url(
    url,
    tool_description_overrides={
        "get_works": "Search Crossref papers by keyword or DOI."
    },
)

Custom Builder Callback

Pass a callback function that receives the Operation model and returns a custom string:

def custom_builder(operation):
    return f"Execute {operation.name} against endpoint {operation.path}."


toolkit = OpenAPIToolkit.from_url(url, description_builder=custom_builder)

---

Operation & Tag Filtering

Filter endpoints before generating LangChain tools to avoid creating unused tools:

Filter by Tags

# Only create tools for the 'Works' tag
toolkit = OpenAPIToolkit.from_url(url, include_tags=["Works"])

# Exclude internal or admin endpoints
toolkit = OpenAPIToolkit.from_url(url, exclude_tags=["Admin"])

Filter by Operations

# Specify exact operations to include
toolkit = OpenAPIToolkit.from_url(url, include_operations=["get_works", "get_members"])

# Exclude destructive operations
toolkit = OpenAPIToolkit.from_url(url, exclude_operations=["delete_work"])