Media

Upload once. Remembered forever.

A workspace-wide media library, not a conversation attachment. Agents auto-load files when they're relevant — that lab panel from three weeks ago surfaces itself when you ask about your supplement plan.

See the "lab panel ghost" painGet started
Media

Upload once. Remembered forever.

Every image, PDF, and document lives in a workspace‑wide media library. Agents decide on their own when a file is relevant and load it — no copy‑paste, no re‑uploading, no “as I mentioned before”.

Learn more
application/pdf
Lab panel — April 2026uploaded 04/29 · #health
labscalprotectinCRP
image/png
Resume v3 — May 2026uploaded 05/02 · #career
resumecareer
application/pdf
Colonoscopy reportuploaded 03/14 · #health
colonoscopyGI
image/heic
Whiteboard — Q3 planuploaded 04/18 · #general
planningstrategy
Recall, six weeks later

You haven’t talked about your labs in a month. You ask Rex about a new symptom. He doesn’t ask for the file. He decides it’s relevant, loads it, and answers grounded in what’s actually in your panel.

From the trace
→ Loaded artifact PDF · Lab panel — April 2026
→ Loaded artifact PDF · Colonoscopy report
→ Loaded profile Health
→ Composing reply…
Scope

Workspace-wide, not conversation-scoped.

Drop a PDF into a thread today and any agent in any channel in that workspace can find it next month. Media isn’t attached to the thread that uploaded it — it’s a workspace-level resource, indexed and addressable by every channel that should see it.

ChatGPT / Claude

Files are tied to a chat or project surface. Recall depends on where you are and what context the model decides to use.

Boardbox media

Uploaded once. Auto-loaded weeks later by any agent in the workspace when relevant.

Title

"Lab panel · April 2026" (auto-derived from content, not the filename).

Description

Two-line summary of what the file contains — read by the index.

Tags

Domain tags ("clinical", "financial", "design"...) that surface the file at the right moment.

On ingest

Vision models pre-process every upload.

The moment you upload a file, a vision-capable model runs over it and generates a title, description, and a small set of tags. Those land in the workspace index so the agent always knows what exists. The file content stays on disk until an agent calls load_media to fetch it for a specific turn.

Recall

Agents pick the right file by name, not by guess.

Because every media item appears in the workspace index (title + description), the agent doesn’t have to guess which file is relevant — it can name the one it needs and call load_media. You see the call inline in the thread as a trace. If the wrong file loaded, you can correct it.

load_media·PDF · Lab panel — April 2026load_media·IMG · Knee MRI · 2026-03-12load_media·PDF · Insurance EOB · Cardiology
Privacy

Encrypted at rest. Workspace-isolated.

Files are stored in encrypted object storage. Workspace boundaries are enforced at the database layer — a query in your Personal workspace cannot reach Professional data, and vice versa. Hard-delete a file and both the binary and its index entry go. We keep nothing in soft-deleted limbo.

FAQ

Common questions

How is this different from uploading a file each time?
Chat apps can attach files, and project spaces can keep some of that context around, but recall is still tied to the chat/project surface and to what the model decides to load. A file uploaded to Boardbox lives in the workspace media library. Agents in any channel of that workspace can auto-load it when it's relevant — that lab panel from three weeks ago surfaces itself when you ask about your supplement plan.
Does Boardbox store my files?
Yes — uploaded files are stored in encrypted object storage scoped to your workspace. You can hard-delete any file from the Media view, which removes both the binary and its index entry. Workspace isolation is enforced at the database layer.
What file formats are supported?
Images (PNG/JPG/HEIC/WebP), PDFs, text files (Markdown, plain text, source code), CSVs, and most common document formats. Vision-capable models can read images directly; PDFs and docs are extracted to text for retrieval.
How does the AI decide when to load a file?
Every uploaded file is auto-titled and tagged on ingest by a vision model. The titles + descriptions live in the workspace index that agents read every turn. When a thread's context calls for a file (e.g. you ask about 'my April labs'), the agent loads it by name via the load_media tool.
Can I share media across workspaces?
No — media is hard-scoped to a single workspace. Your professional design files never appear in your personal workspace. Sharing across workspaces would break the isolation guarantee, so we don't allow it.
See also

Related features

Indexes

Every upload is auto-titled and listed in the workspace index — that's how agents know which file to load.

Learn more

Workspaces

Media is workspace-scoped. Personal media stays personal; professional media stays professional.

Learn more

Integrations

Pull media from Google Drive, GitHub, Notion via MCP — the same auto-load semantics apply.

Learn more

The chatbox was a phase.
Build a workspace.

Boardbox keeps your context, holds its ground, and produces work that lasts.

Get started →