file: README.md

Robodog Code

Overview

Robodog Code is a lightweight, zero-install, fast, command-line style generative AI client that integrates multiple providers (OpenAI, OpenRouter, LlamaAI, DeepSeek, Anthropic, Sarvam AI, Google Search API, and more) into a unified interface. Key capabilities include:

NEVER TRUST A CODE SPEWING ROBOT!

• Access to cutting-edge models: o4-mini (200k context), gpt-4, gpt-4-turbo, dall-e-3, Llama3-70b, Claude Opus/Sonnet, Mistral, Sarvam-M, Gemma 3n, etc.
• Massive context windows (up to 200k tokens) across different models.
• Seamless chat history & knowledge management with stashes and snapshots.
• File import/export (text, Markdown, code, PDF, images via OCR).
• In-chat file inclusion from a local MCP server.
• Built-in web search integration.
• Image generation & OCR pipelines.
• Limit scope of the context window using filter tagging pattern=robodog.py recursive
• AI-driven web automation/testing via Playwright (/play).
• Raw MCP operations (/mcp).
• /todo feature: automate and track tasks defined in todo.md.
• Accessible, retro “console” UI with customizable themes and responsive design.

Try Robodog

• Web: https://adourish.github.io/robodog/robodog/dist/
• Android: https://play.google.com/store/apps/details?id=com.unclebulgaria.robodog
• npm packages:
  • npm install robodoglib
  • npm install robodogcli
  • npm install robodog
• Python:
  • pip install robodogcli
  • pip show -f robodogcli
  • python -m robodogcli.cli --help
  • pip install --upgrade requests tiktoken PyYAML openai playwright pydantic langchain (optional)

Configuration

Click the ⚙️ icon in the top-menu to open settings, or edit your YAML directly:

configs:
  providers:
    - provider: openAI
      baseUrl: "https://api.openai.com"
      apiKey: "<open ai token>"
      httpReferer: "https://adourish.github.io"
    - provider: openRouter
      baseUrl: "https://openrouter.ai/api/v1"
      apiKey: "<open router token>"
      httpReferer: "https://adourish.github.io"
    - provider: searchAPI
      baseUrl: "https://google-search74.p.rapidapi.com"
      apiKey: "<search token>"
      httpReferer: "https://adourish.github.io"
  
  mcpServer:
    baseUrl: "http://localhost:2500"  
    apiKey:   "testtoken"  

  specialists:
    - specialist: nlp
      resume: natural language processing, chatbots, content generation, language translation
    - specialist: gi
      resume: generates images from textual descriptions. understanding and interpreting textual descriptions 
    - specialist: search
      resume: generate simple search results

  models:
    - provider: openRouter
      model: openai/gpt-5-mini
      stream: true
      specialist: nlp
      about: "Best for performance. Context window: 1.05M tokens. Competitive in Academia (#2), Marketing/Seo (#3), Health (#4), Legal (#4), Science (#4)."
    - provider: openRouter
      model: GPT-4o-mini
      stream: true
      specialist: nlp
      about: "Best for most questions. Context window: 1.05M tokens. Pricing: $0.40/M input, $1.60/M output."
    - provider: openAI
      model: o4-mini
      stream: true
      specialist: nlp
      about: "Biggest model with 200k context window and world view. Best for critical thinking. Context window: 200K tokens."
    - provider: openAI
      model: o1
      stream: true
      specialist: nlp
      about: "Big model with 128k context window and small world view. Good for critical thinking. Context window: 128K tokens."
    - provider: openRouter
      model: openai/o4-mini
      stream: true
      specialist: nlp
      about: "Best for big content. Context window: 200K tokens."
    - provider: openRouter
      model: deepseek/deepseek-r1
      stream: true
      specialist: nlp
      about: "Best for summarizing. Context window: 128K tokens. Model size: 671B parameters (37B active). Performance: #2 in Roleplay, #6 in Translation, #9 in Programming, #10 in Science. Supports thinking and non-thinking modes."
    - provider: openRouter
      model: google/gemini-2.5-pro
      stream: true
      specialist: nlp
      about: "Best for speed. Context window: 1.05M tokens. Performance: #3 in Health, #5 in Marketing, Roleplay, Academia, Science. Advanced reasoning, coding, mathematics, scientific tasks. Pricing: $1.25/M input, $10/M output."
    - provider: openRouter
      model: qwen/qwen3-coder
      stream: true
      specialist: nlp
      about: "Best for large docs when speed is not an issue. Context window: 262K tokens. Model size: 480B parameters (35B active). Optimized for agentic coding tasks. Performance: #3 in Programming, #7 in Technology, #8 in Science. Pricing: $0.20/M input, $0.80/M output."
    - provider: openRouter
      model: anthropic/claude-sonnet-4
      stream: false
      specialist: gi
      about: "Best for creating images."
    - provider: openRouter
      model: x-ai/grok-code-fast-1
      stream: false
      specialist: search
      about: "Best for searching. Context window: 256K tokens. Performance: #1 in Programming, #3 in Technology, #6 in Marketing/Seo, #10 in Trivia. Speedy and economical reasoning model. Pricing: $0.20/M input, $1.50/M output."
    - provider: searchAPI
      model: search
      stream: false
      specialist: search
      about: "Best for searching. Context window: 256K tokens. Performance: #1 in Programming, #3 in Technology, #6 in Marketing/Seo, #10 in Trivia. Speedy and economical reasoning model. Pricing: $0.20/M input, $1.50/M output."

Supported Models

OpenAI

• gpt-4, gpt-4-turbo, gpt-3.5-turbo, gpt-3.5-turbo-16k, o4-mini, o1
• dall-e-3

Others

• LlamaAI: llama3-70b
• Anthropic: Claude Opus 4, Claude Sonnet 4
• DeepSeek R1
• Mistral Medium 3, Devstral-Small
• Sarvam-M
• Google Gemma 3n E4B

Key Features

• Multi-Provider Support: Switch between any configured provider or model on the fly (/model).
• Chat & Knowledge: Separate panes for Chat History (💭) and Knowledge (📝)—both resizable.
• Stash Management:
  • /stash <name> — save current chat+knowledge
  • /pop <name> — restore a stash
  • /list — list all stashes
• File Import/Export:
  • /import <glob> — import files (.md, .js, .py, .pdf, images via OCR)
  • /export <file> — export chat+knowledge snapshot
• MCP File Inclusion:
  • /include all
  • /include file=README.md
  • /include pattern=*.js|*.css recursive
  • /include dir=src pattern=*.py recursive
• Raw MCP Operations:
  • /mcp OP [JSON] — e.g. /mcp LIST_FILES, /mcp READ_FILE {"path":"./foo.py"}
• Web Fetch & Automation:
  • /curl [--no-headless] <url> [<url2>|<js>] — fetch pages or run JS
  • /play <instructions> — run AI-driven Playwright tests end-to-end
• Web Search:
  • Use search model or click 🔎 to perform live web queries.
• Image Generation & OCR: Ask questions to dall-e-3 or drop an image to extract text via OCR.
• Interactive Console UI: Retro “pip-boy green” theme, responsive on desktop/mobile, accessible.
• Performance & Size Indicators: Emoji feedback for processing speed and token usage.
• Extensive Command Palette: /help lists all commands, indicators, and settings.
• Todo Automation: Use /todo to execute tasks defined in todo.md across your project roots.

Usage Examples

1) AI-Driven Web Tests with /play

/play navigate to https://example.com, extract the page title, and verify it contains 'Example Domain'

2) Fetch & Scrape with /curl

/curl https://example.com

3) Include Local Files via MCP

/include pattern=*.js recursive fix bug in parser

4) Raw MCP Commands

/mcp LIST_FILES
/mcp READ_FILE {"path":"./src/cli.py"}

5) Switch Model on the Fly

/model o4-mini

6) Import & Export

/import **/*.md
/export conversation_snapshot.txt

7) Auto Side by Side Diff

/todo Feature

Robodog’s /todo command scans one or more todo.md files in your configured project roots, detects tasks marked [ ][-], transitions them to [~][-] (Doing) when started, and [x][-] (Done) when completed. Additionally, flipping from [x[[ ] will commit the changes to from the out file to the destination file(s). Each task may include:

• [ ][-] task status and task commit status
  • include: pattern or file specification to gather relevant knowledge
  • out: file path where the AI will write or update content
  • Optional code fences below the task as initial context

You can have multiple todo.md files anywhere under your roots. /todo processes the earliest outstanding task, runs the AI with gathered knowledge, updates the focus file, stamps start/completion times, and advances to the next.

Example todo.md File Formats

# file: project1/todo.md
- [ ][-] Revise API client
  - include: pattern=api/*.js recursive
  - out: temp/out.js
```knowledge
// existing stub

- [ ][-] Add unit tests
  - include: file=tests/template.spec.js
  - out: temp/out.js

# file: project2/docs/todo.md
- [ ][-] Update README
  - out: file=README.md
- [ ][-] Generate changelog
  - include: pattern=CHANGELOG*.md
  - out: out.md
```knowledge

# todo readme
- [x][-] readme
  - include: pattern=*robodog*.md|*robodog*.py|*todo.md   recursive`
  - out: temp/out.js
```knowledge
1. do not remove any content
2. add a new readme section for the /todo feature with examples of the todo.md files and how you can have as many as possible
3. give lots of exampkes of file formats

todo

• [~][-] changes todo
  • started: 2025-09-16 22:53 | knowledge: 36 | include: 25181 | prompt: 25492 | cur_model: openai/o4-mini
  • include: pattern=robodogclirobodogservice.py|robodogclirobodogtodo.py|robodogclirobodogbuilder.py|robodogclirobodogcli.py recursive`
  • out: temp\out.py

1. detect if the parsed file is new or not. 
2. # file: <filename.ext> NEW
3. add if the file is new to the list of objects from parse_llm_output
4. give me all of the code. 

# watch
- [ ][-] change app prints in service logger.INFO
  - include: pattern=*robodog*.md|*robodog*.py  recursive`
  - out: temp/out.js
```knowledge
do not remove any features.
give me full drop in code file

# fix logging
- [ ][-] ask: fix logging. change logging so that it gets log level through command line. change logger so that it takes log level from the command line param
  - include: pattern=*robodog*.md|*robodog*.py  recursive`
  - focus: file=c:\projects\robodog\robodogcli\robodog\cli3.py
```knowledge
my knowledge

You can chain as many tasks and files as needed. Each can reside in different directories, and Robodog will locate all todo.md files automatically.

Configuration & Command Reference

Robodog UI

See command palette in-app (/help) or the reference below:

/help             — show help  
/models           — list configured models  
/model <name>     — switch model  
/import <glob>    — import files into knowledge  
/export <file>    — export snapshot  
/clear            — clear chat & knowledge  
/stash <name>     — stash state  
/pop <name>       — restore stash  
/list             — list stashes  
/temperature <n>  — set temperature  
/top_p <n>        — set top_p  
/max_tokens <n>   — set max_tokens  
/frequency_penalty <n> — set frequency_penalty  
/presence_penalty <n>  — set presence_penalty  
/dark             - toggle light/dark 
/folders <dirs>   — set MCP roots  
/include …        — include files via MCP  
/curl …           — fetch pages / run JS  
/play …           — AI-driven Playwright tests  
/mcp …            — invoke raw MCP operation  
/todo             — run next To Do task  

Robodog CLI

/help             — show help  
/models           — list configured models  
/model <name>     — switch model  
/clear            — clear chat & knowledge  
/temperature <n>  — set temperature  
/folders <dirs>   — set MCP roots  
/include …        — include files via MCP  

Build & Run

# Clone or unzip robodog
cd robodog
python build.py
open ./dist/robodog.html

npm install robodoglib  
npm install robodogcli  
npm install robodog  
pip install robodogcli  
pip show -f robodogcli  
python -m robodogcli.cli --help  
python -m robodogcli.cli --folders "c:\projects\robodog" --port 2500 --token testtoken --config config.yaml --model  openai/o4-mini --backupFolder "c:\temp"

Enjoy Robodog AI—the future of fast, contextual, and extensible AI interaction!