The agent

Intabula's built-in AI agent works on your vault through a fixed set of typed operations — no shell, no direct file access. It runs on the Claude or Codex subscription you already have: sign in once and it's ready, with no API keys and no per-token billing.

Providers

One agent panel supports two providers: Anthropic's Claude and OpenAI's Codex. Both get the same toolbox and the same boundaries; you switch between them in the panel. Usage counts against the plan you already pay for.

Vault operations

These are the only ways the agent can touch your vault:

Operation	What it does
`list_collections`	List the collections in the vault.
`get_schema`	Read a collection's pinned fields and types.
`query_records`	Query a collection's records with filters and sorting.
`search_vault`	Full-text search across titles, fields, and bodies (accent-insensitive).
`get_record`	Read one record — fields and body.
`get_backlinks`	List the records that link to a given record.
`list_needs_attention`	Read the needs-attention inbox (schema drift, broken links, parse errors).
`open_record`	Open a record in the app for the user.
`open_collection`	Open a collection in the app for the user.
`pin_field`	Promote a recurring field to an official typed column.
`create_record`	Create a record — frontmatter fields plus Markdown body.
`update_record`	Edit fields or body; the edit splices only the changed segment.
`rename_record`	Rename atomically — every wikilink across the vault is rewritten, under a journal.
`move_record`	Move a record between collections; references follow.
`delete_record`	Delete a record.
`create_view`	Save a table or Kanban view to the sidebar.

Read-only context tools

Beyond the vault, the agent can read — never write — from sources you attach:

read_linked_file, list_linked_files, search_linked_files — files in linked folders.
git_log, git_show, git_diff — the Git history of a linked repository, without touching the working tree.
fetch_url and the provider's web search — pulling facts from the web into records.

What the agent does not have

No Bash, no shell, no generic file read/write/edit tools, and no filesystem access outside the boundaries above. Your files leave your machine only as context sent to the model provider, and only within the scope you've granted.

Approval mode

Optionally, every agent write can require your approval. The diff you review is computed by dry-running the actual write without applying it — you see the exact bytes that would land, not the model's summary of its intent. Unanswered approvals deny themselves after a timeout. Approval mode is off by default and per-session.

The activity feed

Every agent write is logged to an activity feed with a before-and-after snapshot of the affected file, so "what changed, and who changed it" always has an answer you can inspect as a diff.