Uploading Files & Project Documents

Ileen lets you give agents extra context by uploading files. Every uploaded file is analysed and turned into a smart, text-based summary — the original files are never stored. This keeps your project fast and your data private.

Files are used in two ways:

Current message attachments — processed and injected into the current turn only.
Project Documents — summaries persisted on the project and automatically re-injected as context in every future turn (Design Mode and Content Mode).

Attaching files in chat

Every chat window has a paperclip (:paperclip:) button next to the message input.

Click the paperclip — a file picker opens.
Select one or more files.
A preview row appears below the textarea:
- Images show a thumbnail
- Text/code files and documents show a file-type icon
- Click × on any file to remove it before sending
Type your message and click Send.

Supported file types

Category	Extensions
Images	`.png`, `.jpg`, `.jpeg`, `.gif`, `.webp`, `.svg`
Text / code	`.txt`, `.md`, `.html`, `.css`, `.js`, `.ts`, `.jsx`, `.tsx`, `.json`, `.py`, `.java`, `.c`, `.cpp`, `.h`, `.go`, `.rs`, `.yaml`, `.yml`, `.xml`, `.sql`, `.sh`, `.csv`, `.toml`, `.ini`, `.cfg`, `.env`
Documents	`.pdf`, `.doc`, `.docx`, `.xls`, `.xlsx`, `.ppt`, `.pptx`, `.odt`, `.ods`, `.rtf`

How files are analysed

Every file you upload goes through an intelligent analysis tailored to its type. The original file is never kept — only a compact, AI-generated summary.

flowchart LR
    Upload[You upload a file] --> Type{What kind of file?}
    Type -->|Text or code| ST[Read and summarise<br>if very long]
    Type -->|Document| DT[Extract the text<br>then summarise]
    Type -->|Image| IT[Computer vision<br>describes what it sees]
    DT --> Store[Summary saved to project]
    IT --> Store
    ST --> Store

How each file type is handled

File type	What happens
Text / code (`.py`, `.json`, `.md`, …)	Read directly. Short files are kept in full; longer files get a structured summary so the agent gets the key points without the noise.
PDF (`.pdf`)	Text is automatically extracted from all pages and tables, then summarised into a compact overview.
Word (`.docx`, `.doc`)	Text is extracted from paragraphs and tables, then summarised.
Excel (`.xlsx`, `.xls`)	Data is read sheet-by-sheet, then summarised — the agent gets the structure and key facts, not raw cells.
PowerPoint (`.pptx`, `.ppt`)	Text is extracted from every slide, then summarised.
Images (`.png`, `.jpg`, …)	Analysed by computer vision. The agent receives a rich description covering what’s visible, the layout, colours, any text in the image, and the overall design style.

What a document summary looks like

Each document is summarised into a compact, structured JSON object:

{
  "doc_type": "requirements",
  "purpose": "Functional and non-functional requirements for the e-commerce platform.",
  "key_topics": ["user auth", "payment integration", "inventory", "API design"],
  "important_facts": ["Must support Stripe and PayPal", "10k concurrent users", "GDPR"],
  "structure": "7 sections: overview, auth, payments, inventory, API, security, deployment",
  "actionable_items": ["Define auth provider", "Clarify multi-currency requirements"]
}

What an image analysis covers

Images are analysed by computer vision and converted into a detailed text description:

Description — what the image shows, in natural language
UI Elements (for screenshots) — buttons, inputs, tables, cards, navigation, charts, modals, forms…
Color Palette — dominant colours with hex codes and semantic roles (primary, accent, background…)
Layout Structure — grid, flex, centred form, dashboard layout…
Typography — font families, sizes, weights, visual hierarchy
Visible Text (OCR) — any legible text rendered in the image
Design Patterns — hero section, card grid, login form, pricing table…
Visual Tokens — keyword tags (e.g. dashboard, dark-theme, saas, minimalist)

Attachments in Develop Mode

When you’re working with an existing codebase in Develop Mode, you can attach files just like in Design Mode — but the behaviour is tailored for code-focused work.

Where you can attach files

Attachments are available in two Develop Mode agents:

Code Query — attach screenshots, specs, or reference docs when asking questions about your codebase
Task Plans (Discuss phase) — share mockups, requirements PDFs, or error screenshots while refining a task plan with the agent

How it helps

Attaching files in Develop Mode is especially powerful for:

Scenario	What to attach	What happens
Bug report	Screenshot of the error	The agent sees the exact error message and stack trace, not just your description
UI review	Mockup or design screenshot	The agent understands the intended layout, colours, and components before writing code
Spec implementation	PDF or Word document with requirements	The agent reads the spec and maps it to the codebase automatically
API integration	Reference `.json` or `.yaml` file	The agent parses the schema and checks if the existing code matches
Refactoring guidance	Architecture diagram screenshot	The agent understands the target architecture and plans accordingly

What makes it different from Design Mode

In Develop Mode, attached files stay within the current conversation session — they’re not saved as permanent project documents. This keeps your developer workspace clean and focused: every session is self-contained with the context you’ve shared in it.

The conversation history preserves the analysis, so if you attach a mockup in your first message and discuss it across five follow-up questions, the agent remembers every detail without you needing to re-attach it.

Project Documents — persistent context (Design & Content Mode)

In Design Mode (Architect chat) and Content Mode (Briefing), uploaded files are automatically persisted as Project Documents. Each file’s summary becomes part of the project’s permanent knowledge base and is re-injected as context in every future agent interaction — until you delete it.

Open the Project Documents panel from the project sidebar (:folder_open:) to manage them.

What’s in the panel

Column	What it shows
Icon	:framed_picture: / :code: / :page_facing_up: based on type
Filename	Original name at upload time
Type badge	`Image` / `JSON` / `Python` / `PDF` / `Word` / `Excel` / `PowerPoint` / `Text`
Size	Original file size (human-readable)

Managing documents

Refresh — click :arrows_counterclockwise: to re-fetch the list
Delete — click :wastebasket: on any document, confirm the prompt, and it’s removed. The document summary will no longer be used as context from the next message onward.
The footer shows the total count: “N documents • All injected as context”

When to use which

Use case	Use
One-off: explain a screenshot, ask about a single PDF	Chat attachment (current message only)
A requirements doc the architect needs for the full design phase	Chat attachment → auto-saved as Project Document
A brand guide that should influence every agent call	Project Document
Screenshots of a competitor’s UI for design inspiration	Chat attachment → image analysed → summary persisted
Company tone-of-voice sample files	Project Document
Outdated specs you want to remove	Delete from the Project Documents panel