🧊 SalticeAgent — Autonomous AI Engineering & Visual Studio

SalticeAgent is a production-grade, autonomous engineering agent and creative studio. Built for multi-turn autonomous coding, surgical self-healing, multi-model LLM orchestration, and integrated visual asset synthesis (diffusion + mathematical vector SVG generation).

---

⚡ Key Highlights

• Global Terminal CLI (saltice): Launch from anywhere in Windows, macOS, or Linux.
• Dual-Mode Resilient Parsing: Robust JSON parser with Markdown block recovery (SalticeAntigravityEngine) — never halts on imperfect LLM outputs.
• Infinite Execution Runway & Self-Healing: Captures compiler/test exit codes, tracebacks, and lint failures; analyzes line slices and applies surgical patches until verified.
• Asynchronous Subagent Swarm: Spawn specialized background worker threads (planner, architect, tdd-guide, security-reviewer, image-agent, vector-designer) that work in parallel.
• Integrated ImageAgent & Vector Studio:
  • Edge Diffusion (<2.5s): Instant serverless generation powered by Cloudflare Workers AI (@cf/black-forest-labs/flux-1-schnell, SDXL).
  • ZeroGPU Diffusion Matrix: Multi-model parallel dispatch across Hugging Face Spaces (FLUX.1-dev, SD 3.5 Large) with Pollinations.ai fallback.
  • 4-Layer Resilient Vector SVG Engine: Mathematical SVG synthesis with <defs> glowing gradients, Iconify registry integration (200k+ icons), curated presets, and lossless PNG rasterization via svglib + reportlab.
  • Diffusion Typography & Wordmarks: Automated PIL font auto-fitting to overlay crisp text on diffusion assets without AI hallucination.
  • C.R.A.P. Design Validation: Autonomous scoring of contrast, repetition, alignment, and proximity.
• Multi-Tier Research Matrix: Automated failover across Tavily AI, Google News RSS, Jina AI Reader, and DuckDuckGo.
• Karpathy & ECC Directives: Built-in adherence to simplicity, surgical edits, and goal-driven verification.

---

🏗 Architecture Overview

graph TD
    User([User Prompt / CLI]) --> Engine[Saltice Autonomous Engine]
    
    subgraph Control Plane
        Engine --> Parser[Dual-Mode Parser: JSON + Markdown Recovery]
        Engine --> Memory[4-Tier Memory & State Manager]
        Engine --> Subagents[Asynchronous Subagent Swarm]
    end

    subgraph LLM Providers
        Engine --> Gemini[Google Gemini 2.5 / 3.1]
        Engine --> Groq[Groq Llama / Qwen]
        Engine --> DeepSeek[DeepSeek V3 / R1]
        Engine --> Local[Local Ollama / LM Studio]
    end

    subgraph Tooling & Execution Matrix
        Engine --> Tools[38+ Unified Tools]
        Tools --> FileOps[Surgical File Operations]
        Tools --> Shell[Non-Blocking Process Runner]
        Tools --> Search[4-Tier Search: Tavily / Jina / DDG]
        Tools --> Browser[Chrome DevTools MCP / Automation]
    end

    subgraph ImageAgent Studio
        Engine --> VisualStudio[Integrated ImageAgent]
        VisualStudio --> CF[Cloudflare Workers AI: Flux Schnell <2.5s]
        VisualStudio --> HF[HuggingFace ZeroGPU: Flux Dev / SD 3.5]
        VisualStudio --> SVG[4-Layer Vector SVG Engine]
        SVG --> svglib[svglib + reportlab Rasterizer]
        VisualStudio --> Wordmark[PIL Auto-Fitted Typography]
    end

Loading

---

🚀 Quickstart & Installation

1. Clone the Repository

git clone https://github.com/The-Do-ers/SalticeAgent.git
cd SalticeAgent

2. Set Up Virtual Environment & Dependencies

python -m venv .venv

# On Windows (PowerShell)
.\.venv\Scripts\Activate.ps1

# On Linux / macOS
source .venv/bin/activate

# Install requirements
pip install -r requirements.txt

3. Configure Environment Variables

Copy .env.example to .env and fill in your API keys:

cp .env.example .env

(At minimum, configure GEMINI_API_KEY or GROQ_API_KEY for LLM operations. Add CLOUDFLARE_ACCOUNT_ID and CLOUDFLARE_API_TOKEN for sub-2.5s image generation).

4. Install Global CLI Command (Optional)

pip install -e .

Now you can run saltice from any directory!

---

🖥 Usage

Interactive Shell

saltice

Starts the interactive agent session with live streaming, syntax highlighting, and autonomous execution.

Single Prompt Mode

saltice "Build a FastAPI REST service with JWT auth and write pytest test cases"

Visual & Logo Generation CLI

You can also invoke the image_agent module directly:

# Generate a fast diffusion image with Cloudflare Flux Schnell
python -m image_agent.cli --prompt "Cybernetic cityscape at sunset" --output artifacts/city.png

# Generate a high-resolution mathematical vector SVG logo
python -m image_agent.cli --logo "Quantum security shield with cyan gradients" --style modern-tech --output artifacts/shield.svg

---

🛠 Unified Tool Suite (Highlights)

Category	Tools	Description
Filesystem	read_file, write_file, replace_file_content, view_file_slice	Surgical, non-destructive file reading, creation, and line-level diff patching.
Execution	run_command, run_tests	Non-blocking execution of tests, compilers, and linters with exit code analysis.
Web & Research	search_web, fetch_webpage, tavily_search	Multi-provider fallback web scraping and deep synthesis.
Subagents	spawn_subagent, check_subagent_status, list_subagents	Asynchronous multi-threading for delegated tasks.
Visual Studio	generate_image, generate_svg_logo, inspect_image_quality	Parallel diffusion, SVG vector design, typography typesetting, and image quality checks.
Inspection	get_workspace_summary, view_file_outline	AST-aware code navigation and directory mapping.

---

👥 Subagents & Modular Skills

SalticeAgent includes pre-configured specialized personas in subagents.py:

• planner: Decomposes complex projects into testable milestones.
• architect: Enforces low-coupling module design and clear boundaries.
• tdd-guide: Writes unit and integration tests before implementation.
• code-reviewer: Reviews security, performance, and anti-regression.
• image-agent: Autonomous visual studio for diffusion mockups and UI assets.
• vector-designer: Mathematical SVG geometry and brand mark synthesizer.

Skill modules are loaded dynamically from skills/:

• creative-web-design: Awwwards-grade layouts, typography, and visual patterns.
• gsap-lenis-motion: Smooth scroll and timeline animations.
• kokonut-ui & react-bits: Modern interactive component libraries.
• test-driven-development: Pytest workflows and verification gates.
• image-agent: Visual asset and vector generation blueprints.

---

🔐 Security & Best Practices

• Never commit .env: SalticeAgent includes a strict .gitignore that ignores .env, runtime databases (telemetry.db), session logs, and cache files.
• Non-blocking Dev Servers: As per the engineering directives, SalticeAgent verifies code using non-blocking commands (npm run build, pytest, python -m unittest) and avoids hanging background servers.
• Surgical Edits: Edits are performed using replace_file_content targeting exact anchors to avoid wiping existing implementations.

---

🤝 The Do'ers Organization

Maintained by The-Do-ers team.

• Organization: https://github.com/The-Do-ers
• Contributing: Fork the repository, create a feature branch, and submit a pull request.