Kit for AI

Convert Url

Fetch a document or web page by URL and convert it to clean Markdown (or structured JSON). Supports PDF, Office files, HTML, CSV, and images.

List Recent Conversions

List the user's most recent conversions (id, file name, status).

List Knowledge Bases

List the user's knowledge bases (named knowledge bases of converted documents), with each knowledge base's id, name, kind, and document count.

Read Knowledge Base

Read every document in a knowledge base as one combined Markdown document — use this to give the model access to all docs in a category. Large knowledge bases are truncated to bound token usage; prefer search_knowledge_base for targeted retrieval. Reference the knowledge base by id or name.

Search Knowledge Base

Semantically search a knowledge base and get back the most relevant chunks (ranked by score) instead of the whole knowledge base — use this for large knowledge bases or targeted questions. Reference the knowledge base by id or name.

Write Knowledge Base

Write content back into a knowledge base as a versioned document — use this to save notes, memories, or generated content the user can review and revert. Creates a new document, or replaces an existing one's content (a new revision) when you pass its title as the document. Reference the knowledge base by id or name.

Append Knowledge Base

Append content to a knowledge base — adds a new versioned document with the given content (searchable like any other doc). Reference the knowledge base by id or name.

Edit Knowledge Base

Correct an existing document in a knowledge base by replacing a specific piece of text — use this for targeted fixes instead of rewriting the whole document. Creates a new revision (the old one is kept, revertible). `find` must be text that currently appears in the document. Get the document id from search_knowledge_base or list results.

Remember

Save a fact, preference, or piece of context to the user's persistent memory. Use after learning something worth keeping across conversations (user preferences, project facts, decisions). Near-duplicate memories are automatically skipped. Stored in the auto-created 'Memory' knowledge base, reviewable by the user.

Remember Conversation

Extract and store the durable facts from a conversation transcript in ONE call — preferences, decisions, stable facts (transient chit-chat is dropped, near-duplicates skipped). Use at the end of a session instead of many individual remember calls.

Memory Profile

Get a compact synthesized profile of the user (or one end user) from their stored memories — who they are, preferences, projects, decisions in force. Cheaper than many recall calls when you just need orientation at the start of a session.

Recall

Search the user's persistent memory for relevant facts/context before answering. Returns the most relevant remembered items with scores. Use at the start of tasks to recover prior context.

Crawl

Crawl a website (same-domain, bounded) and build a searchable knowledge base from it — then your agent can search_knowledge_base it. Give a start URL and a knowledge base name.