# Attachments — images and files in a message Attach an image or a text file to a message when the model can read it: what's supported, how the app gates the button on the model's own capabilities, and where the bytes go. Status: Shipped An attachment is an image or a text file you send alongside your message. The model reads it as part of the turn — describing a screenshot, reviewing a log, answering questions about a config file. What you can attach depends on the model you brought: the app only offers what the model advertises it can read. ## Why it matters A prompt is text, but the thing you actually want help with often isn't. An attachment lets you hand the model the artifact itself — a photo, an error log, a CSV — instead of retyping or describing it. And because the attach options are driven by the model's advertised capabilities, the app never lets you send an image to a model that would reject the whole turn for it. ## What you can attach - **Images**, when the model advertises image input (vision): PNG, JPEG, GIF, WebP, BMP, AVIF, and SVG. - **Text files**, when the model advertises file input: plain text and log files, source code in many languages, Markdown, JSON / JSONL, CSV / TSV, XML, YAML, HTML, and CSS. Source-code files of every kind are read as plain text; the model still sees the real filename, so it knows it's looking at, say, a Dart file rather than a Python one. ## Attaching one The composer's attach button opens a short menu. Which entries appear depends on what the model accepts: - **Add image…** / **Add file…** — pick from your device. - **Take a photo** — capture with the camera. Phones only; desktop has no camera to reach for. - **Paste from clipboard** — drop in an image you copied elsewhere. You can also insert an image straight from a keyboard that supports it. Each attachment shows in the composer before you send — an image as a thumbnail, a file as a document chip — and you can remove any of them before sending. The attach button itself is hidden entirely when the model accepts neither images nor files, so it's never a dead control. ## How the model receives it - An **image** goes to the model inline with your message. - A **file** goes as a native file part where the backend accepts one, and is otherwise folded into the message text so its contents still reach the model — either way the model reads the file. If you switch a conversation to a model that _can't_ read an image already in it, the turn still runs: the image is left out of the request and the app shows a warning saying so, rather than failing the send. The point is that a model swap never silently breaks a turn. An on-device vision model reads images through its projector file. When a reply was written without the conversation's images — the model reads text only, its projector isn't on this device, or the projector failed to load — a note under the reply says how many images it missed and why, and a projector that failed to load can be tried again from there. See [On-device models](/docs/features/embedded-models#when-the-vision-projector-loads). ## Where attachments live Attachment bytes are stored **encrypted on your device**, in a blob store separate from the conversation database, and are cleaned up when you delete the message or conversation that referenced them. This is the same local-first, encrypted-at-rest handling the rest of your data gets — see [Local-first & private by design](/docs/features/local-first). ## What it doesn't do yet - **Images and text files only.** There's no PDF, audio, or video attachment path today — a PDF is not among the pickable file types. - **The model must advertise the capability.** If a provider doesn't report a model's input modalities, the app can't know it accepts images or files, so the attach options for those stay hidden even if the model would in fact accept them. Bring a provider that advertises what its models can read. More in [Features](/docs/features): [Providers — bring your own LLM](/docs/features/providers) · [On-device models — run fully offline](/docs/features/embedded-models) · [Agents — the thing you build](/docs/features/agents) · [Tools — giving an agent more than the model knows](/docs/features/tools) · [Workspaces — keep your work in scopes](/docs/features/workspaces) · [Conversations — how a turn runs](/docs/features/conversations) · [Context management — what the model still sees](/docs/features/context-management) · [Background turns — replies that keep running](/docs/features/background-turns) · [Local-first & private by design](/docs/features/local-first) · [Sync — your data across your devices](/docs/features/sync) · [Sharing — move agents, tools and providers as a file](/docs/features/sharing) · [Marketplace — find and publish shared agents, tools and providers](/docs/features/marketplace) · [Running a Couchbase Edge Server](/docs/features/couchbase-edge-server) · [Spell checking — catch typos while you write](/docs/features/spell-check) --- [Bring Your LM](/) · [Docs](/docs/features) · [Marketplace](/marketplace) [Articles](/articles) · [Releases](/releases) · [Bundle Format](/marketplace/bundle-format) · [Marketplace Terms](/terms) · [App Privacy](/privacy) · [Website Privacy](/website-privacy) · [Open Source](/open-source) · [llms.txt](/llms.txt)