LLMs.txt
LLMsTxtTools lets an agent discover and read documentation from an llms.txt index, optionally loading it into Knowledge.
LLMsTxtTools reads llms.txt files. The format is a standardized way for websites to publish an LLM-friendly documentation index. The toolkit operates in two modes depending on whether you pass a Knowledge instance.
Agentic Mode
Without knowledge, the agent reads the index and decides which pages to fetch.
1from kern.agent import Agent2from kern.models.openai import OpenAIResponses3from kern.tools.llms_txt import LLMsTxtTools45agent = Agent(6 model=OpenAIResponses(id="gpt-5.4"),7 tools=[LLMsTxtTools()],8 instructions=[9 "First use get_llms_txt_index to see what pages are available.",10 "Then use read_llms_txt_url to fetch only the pages relevant to the question.",11 ],12 markdown=True,13)1415agent.print_response(16 "Using the llms.txt at https://kern.ndx.rocks/llms.txt, "17 "find and read the documentation about how to create an agent with tools",18 stream=True,19)Knowledge Mode
Pass a Knowledge instance and the toolkit exposes read_llms_txt_and_load_knowledge, which ingests the indexed pages into your knowledge base for retrieval.
1from kern.knowledge.knowledge import Knowledge2from kern.tools.llms_txt import LLMsTxtTools34tools = LLMsTxtTools(knowledge=knowledge)Toolkit Params
| Parameter | Type | Default | Description |
|---|---|---|---|
knowledge | Optional[Knowledge] | None | When set, switches to Knowledge mode (load pages instead of returning). |
max_urls | int | 20 | Maximum number of pages to read from an index. |
timeout | int | 60 | HTTP timeout in seconds. |
skip_optional | bool | False | Skip pages listed under the index's optional section. |
allowed_hosts | Optional[List[str]] | None | SSRF guard. Only fetch from these hosts. None allows any host. |
Toolkit Functions
| Function | Mode | Description |
|---|---|---|
get_llms_txt_index | Agentic | Fetch and parse an llms.txt index. |
read_llms_txt_url | Agentic | Read a single page referenced by the index. |
read_llms_txt_and_load_knowledge | Knowledge | Read indexed pages and load them into Knowledge. |
All functions have sync and async variants.