Multimodal Input

FastAIAgent agents accept images, PDFs, and any file/bytes as first-class inputs alongside text. The same code works against OpenAI, Azure, Anthropic, Gemini, Bedrock, and Ollama — the SDK handles the provider-specific wire formatting and forwards files natively (never stringified, never rendered locally when the provider can parse them).

from fastaiagent import Agent, LLMClient, Image, PDF, File

agent = Agent(
    name="claims",
    llm=LLMClient(provider="anthropic", model="claude-sonnet-4-6"),
)

result = agent.run(
    [
        "Compare the photo to the policy and assess the claim.",
        Image.from_file("damage.jpg"),
        PDF.from_file("policy.pdf"),
    ]
)
print(result.output)

Or just hand the agent bytes / a path — the type is sniffed and wrapped in a File automatically:

agent.run(file_bytes)                       # mime sniffed from the bytes
agent.run(["Summarize this.", Path("q3.csv")])

What's supported

Capability	Status
Image input (JPEG, PNG, GIF, WebP)	Yes
PDF input (text / vision / native modes)	Yes
Native PDF (OpenAI/Azure file, Anthropic/Bedrock document, Gemini inlineData)	Yes
Generic File input (any bytes, mime-routed)	Yes
Auto-detect bare bytes / Path	Yes
Documents (csv/txt/html/md, docx/xlsx via Bedrock/Gemini)	Yes
Audio (wav/mp3 on OpenAI/Gemini)	Yes
Tool returns Image / PDF	Yes
Multimodal in Chain state (checkpointed)	Yes
Multimodal in Swarm shared context	Yes
Multimodal in eval JSONL datasets	Yes
Multimodal in trace inspector + Replay	Yes

Input-only: image/audio generation and OCR-as-a-feature are out of scope. Which file types a given provider accepts natively varies — see the File routing matrix.

Read next

• When to use it — read this first if you're weighing the abstraction against just base64-encoding the file yourself. Honest decision rubric for when the layer earns its keep and when it's overkill.
• Images — Image constructors, supported formats, sizes, the OpenAI detail parameter.
• PDFs — PDF constructors, text vs vision vs native modes.
• Files (any type) — the generic File input, auto-detect, and the per-provider routing matrix.
• Provider Support — capability matrix and auto-detection rules.
• Multimodal Eval — JSONL syntax for image/PDF references.