How does Takibi Base prepare documents for AI agent search?
Takibi Base prepares documents by having you upload files into projects and folders, converting them to Markdown so agents can search them, and tracking each file's state as indexed, converting, or quarantined. Once a file is indexed, it becomes searchable by any Profile whose folder permissions include it. You can also download the original files, re-extract their text, or export a project in Open Knowledge Format (OKF).
The preparation pipeline
The path from an uploaded file to a citable passage has three stages:
- Upload — You add documents into projects and folders. Folders are the unit of organization that Profiles later use for access control.
- Convert — Takibi converts the files to Markdown so agents can search them. Markdown is the searchable representation, not necessarily what you uploaded.
- Index — Once conversion finishes, the document is indexed and available to search.
Folder access includes documents you add later, once they are indexed — so you don't have to re-grant permissions every time you add a file to an existing folder.
File states you can check
Takibi exposes the processing state of each file so you can tell what is actually searchable:
| State | What it means |
|---|---|
| Indexed | Converted and available for agent search |
| Converting | Still being processed; not yet searchable |
| Quarantined | Held back from search |
Checking these states matters because a file that is still converting or quarantined will not appear in results, even if its folder is permitted for a Profile.
Working with originals and exports
Preparation does not lock you into the converted form:
- Download the original files — the source you uploaded remains retrievable.
- Extract their text again — re-run text extraction on a document.
- Export a project in Open Knowledge Format (OKF) — take the project's content out in a portable format.
What agents actually receive
When an agent asks a question through the CLI, Takibi searches the permitted sources and returns matching passages with citations and an evidence support score. For example, asking takibi ask -q "What is our refund policy?" --json returns spans containing the exact passage text, a documentId, documentName, chunkId, and title, plus a citations array and a support score. If it finds no matching evidence, it returns no passages.
This is why the conversion and indexing steps matter: the passage an agent cites comes from the Markdown representation of your document, and only indexed documents can produce those passages.
Practical notes
- Uploads, imports, downloads, and billing require your own workspace. The sample workspace lets you try the flow, but edits reset on refresh.
- Because folder access extends to documents added later once indexed, plan your folder structure around which tools should see which material, then let indexing bring new files into scope automatically.
- Project-level access is broader than folder-level access: it also includes future folders and documents outside folders, and Takibi asks you to confirm that broader access.