File attachments

File attachments

Ptah uses a single @ autocomplete — the unified suggestions dropdown — to attach anything in your workspace as context for the next message.

Opening the dropdown

Type @ in the composer. The dropdown opens and filters as you type. Results are grouped by type:

Group	What it is	Typical use
Files	Any file tracked by the workspace.	Attach source code, config, or docs as context.
Folders	Directories in the workspace.	Give the model a whole module or package.
Agents	Specialist agents from .claude/agents/ or the built-in catalog.	Delegate work to a domain expert.
Skills	Skills from .claude/skills/ or the harness registry.	Pull in procedural knowledge for a task.
MCP resources	Anything exposed by a connected MCP server.	Attach live tickets, PRs, or search results.

Use Arrow keys to move between results and Enter or Tab to insert. Esc closes the dropdown.

How attachments are sent

• Files are read at send-time and inserted into the prompt as fenced code blocks with the file path as a header. The model sees the current on-disk contents, not a stale snapshot.
• Folders expand to a file tree plus the contents of every text file inside, subject to a size budget. Binary files are skipped.
• Agents become delegation targets — the model can hand a subtask to the agent with ptah_agent_spawn.
• Skills are injected as procedural context at the top of the system prompt.

Tip

If you attach a large folder, Ptah shows the projected token count before you send. Use the Execution Tree after the turn to see exactly how many tokens each attachment consumed.

Drag and drop

You can drop files from your OS file manager directly onto the composer. They are attached the same way as @-mentions. Images (.png, .jpg, .gif, .webp) are sent inline to vision-capable models.

Removing attachments

Attachments render as chips above the composer. Click the x on a chip to remove it before sending. Once a message has been sent, its attachments are frozen into the transcript.