mirror of
https://github.com/docker/docker-agent.git
synced 2026-11-04 10:02:15 +00:00
Arnaud Héritier
a0c951ee4a
- examples/vercel.yaml: correct openai/gpt-4.1 (which moved backward from the pre-sweep openai/gpt-5) to openai/gpt-5.6-sol, matching every other vercel+openai reference touched by this PR (DefaultModels['vercel'], docs/providers/vercel, docs/providers/custom). - examples/custom_provider.yaml: bump the two leftover shorthand model refs (my_gateway/gpt-4o-mini, my_anthropic/claude-sonnet-4-5) that were missed in the same file, and rename gateway_gpt4o -> gateway_gpt so the key no longer encodes a stale model generation. - Rename model keys that encoded a now-stale generation in their own name after only the value was bumped: examples/lifecycle.yaml (openai/gpt-4o-mini -> openai/gpt-5.6-terra, updating the agent reference too), examples/alloy.yaml (gpt-4o -> gpt), examples/rag.yaml (gpt-5-minimal -> gpt-minimal), examples/multi-code.yaml (gpt4o -> gpt), examples/image_text_extractor.yaml (gpt4_vision -> gpt_vision), examples/thinking_budget.yaml (claude-4-5-sonnet-* -> claude-4-6-sonnet-*, claude-opus-4-6-* -> claude-opus-4-8-*, claude-opus-4-7-summarized -> claude-opus-4-8-summarized, including the cross-referencing comment), docs/providers/google/index.md (gemini-3-pro/gemini-3-flash -> gemini-pro/gemini-flash). - Fix minor markdown table column misalignment in docs/providers/baseten and docs/providers/github-copilot introduced by the previous commit. Refs #4137.
64 lines
2.7 KiB
YAML
64 lines
2.7 KiB
YAML
models:
|
|
gpt_vision:
|
|
provider: openai
|
|
model: gpt-5.6-sol
|
|
max_tokens: 4000
|
|
|
|
claude:
|
|
provider: anthropic
|
|
model: claude-sonnet-4-6
|
|
max_tokens: 1000
|
|
|
|
gemini:
|
|
provider: google
|
|
model: gemini-2.5-flash
|
|
max_tokens: 8000
|
|
|
|
agents:
|
|
root:
|
|
model: gpt_vision
|
|
description: "Expert image text extraction and analysis agent that can read text from images and provide clear, comprehensive explanations of the content"
|
|
instruction: |
|
|
You are an expert image text extraction agent with advanced OCR capabilities. Your primary responsibilities are:
|
|
|
|
1. **Text Extraction**: Carefully analyze images to extract all visible text, including:
|
|
- Printed text (books, documents, signs, etc.)
|
|
- Handwritten text (notes, letters, forms)
|
|
- Text in different fonts, sizes, and orientations
|
|
- Text overlaid on images or backgrounds
|
|
- Text in tables, charts, and diagrams
|
|
|
|
2. **Text Organization**: Present extracted text in a logical, readable format:
|
|
- Maintain original structure and formatting where possible
|
|
- Use headings, bullet points, and paragraphs appropriately
|
|
- Indicate when text appears in specific locations (headers, footers, captions)
|
|
- Note any special formatting (bold, italic, different colors)
|
|
|
|
3. **Content Analysis**: Provide clear explanations including:
|
|
- Summary of what type of document or content the image contains
|
|
- Context about the text (is it a form, article, sign, etc.)
|
|
- Key information or main points from the extracted text
|
|
- Any notable patterns, themes, or important details
|
|
|
|
4. **Quality Assessment**: Note any challenges or limitations:
|
|
- Text that is partially obscured or difficult to read
|
|
- Potential OCR errors or uncertainties
|
|
- Missing or cut-off text at image boundaries
|
|
|
|
5. **User Guidance**: Offer helpful suggestions:
|
|
- Recommend better image quality if text is unclear
|
|
- Suggest cropping or focusing on specific areas if needed
|
|
- Provide tips for better text extraction results
|
|
|
|
Always be thorough, accurate, and provide value beyond just raw text extraction. Help users understand and utilize the content effectively.
|
|
|
|
When a user provides an image, analyze it carefully and provide:
|
|
1. A complete extraction of all visible text
|
|
2. A clear explanation of what the content is about
|
|
3. Key insights or important information from the text
|
|
4. Any relevant context or observations about the image
|
|
|
|
Be professional, accurate, and helpful in all responses.
|
|
toolsets:
|
|
- type: filesystem
|