🦙 Ollama Workbench - Local AI Workbench

Looking for the next generation? Check out Ollama Workbench 2.0 — a SvelteKit + Tauri rewrite with a native desktop app, MCP Studio, and visual workflow builder.

Ollama Workbench is a comprehensive platform for managing, testing, and using AI models from the Ollama library and external providers (OpenAI, Groq, Mistral). It bundles chat (text, voice, vision), retrieval-augmented generation, multi-agent workflows, model management and benchmarking, and an OpenAI-compatible API - all running locally against your own Ollama server.

🚀 Quick Start

One-Command Setup:

git clone https://github.com/marc-shade/Ollama-Workbench.git
cd Ollama-Workbench
python3 scripts/setup_workbench.py

Start the Platform:

./start_workbench.sh  # Unix/Linux/macOS

Access the Interface:

Web UI: http://localhost:8501
OpenAI-compatible API: http://localhost:8000/v1 (starts with the app; toggle with ENABLE_OPENAI_COMPAT)
TTS server (optional, standalone): http://localhost:5002 (tts_server/start_tts_server.sh)
Default login: No authentication required initially (configurable)

✨ What's New (July 2026)

Modern dark UI built on Streamlit's native theming (.streamlit/config.toml) — orange accent, themed sidebar, rounded widgets; customize the whole look by editing one file
OpenAI-compatible API server starts with the app on port 8000, so any OpenAI client can talk to your local Ollama models
Hardened live paths: chat streaming, voice chat, RAG fallback, workflows, and telemetry all repaired and verified against a running Ollama server
1,076-test suite, fully green — hermetic (no live server required), run it with venv/bin/python -m pytest tests/

🌟 Key Features

🏢 Solid Foundations

Encrypted API-key storage (AES-256-GCM) and integrity-hashed audit logging
Security toolkit included: JWT auth, RBAC, and encryption modules ready to wire up (see Security Features for what's active by default)
Observability with optional Opik integration and local performance monitoring
Streamlit UI + Flask API services with an OpenAI-compatible endpoint

🤖 Multi-Provider AI Integration

Ollama Models - Local LLM execution with full privacy
OpenAI API - GPT-3.5, GPT-4, and latest models
Groq - Ultra-fast inference with Llama and Mixtral
Mistral AI - European AI excellence
Unified Interface - Single platform for all providers

🔄 Advanced Workflow Orchestration

Multi-Agent Systems with specialized AI agents
Visual Workflow Builder (Nodes) for complex pipelines
Project Management with AI-assisted planning
Brainstorming Sessions with collaborative AI teams
Research Automation with multi-source intelligence

🛡️ Security Features

An honest map of what ships in the security/ package: what runs by default, what is included but must be wired up, and what is not implemented.

✅ Active by default

Encrypted API-key storage - provider keys are encrypted at rest with AES-256-GCM (cryptography library) when ENABLE_ENCRYPTION is on
Security audit logging - configuration changes and security events are written as structured audit records, each with a SHA-256 integrity hash (ENABLE_AUDIT_LOGGING)
Security framework self-check - configuration validation plus a self-assessment report that scores the current settings against GDPR / HIPAA / SOX / ISO27001 checklists (a gap report, not a compliance certification)

🧰 Included, but not wired into the UI yet

The app currently runs without authentication - these modules exist in security/ and are importable, but no login gate protects the Streamlit UI:

JWT authentication (security/authentication.py) - bcrypt password hashing, JWT sessions, and account lockout after configurable failed attempts. The Streamlit auth UI hook exists but is disabled by default.
Role-Based Access Control (security/access_control.py) - role and permission model (guest/user/admin) ready to enforce, currently not called by any UI path
Key rotation (security/encryption.py, security/security_config.py) - rotate_keys() / rotate_secrets() methods with a rotation-interval setting; rotation is manual (nothing schedules it automatically)

❌ Not implemented (do not rely on these)

Multi-factor authentication - the user model has MFA fields, but there is no TOTP generation or verification code
Transport encryption - the app serves plain HTTP on localhost; put a TLS proxy in front for remote access
Tamper-proof logs - audit events carry integrity hashes but are not chained or signed
Virus scanning of uploads, intrusion detection, data-loss prevention

Deployment guidance: bind to localhost (the default), treat the app as a single-user local tool, and wire up the auth module before exposing it to a network you do not control.

📊 Observability Features

📈 Performance Monitoring

Real-time Metrics with custom dashboards
Performance Alerts with configurable thresholds
Token Usage Tracking across all providers
Response Time Analysis with percentile breakdowns
Resource Utilization monitoring (CPU, memory, GPU)

🔍 LLM Operation Tracing

Opik Integration for comprehensive LLM observability
Request/Response Tracing with full context capture
Error Tracking with detailed error analysis
Cost Analytics across different AI providers
Usage Pattern Analysis for optimization insights

📊 Analytics & Reporting

Usage Reports with trend analysis
Model Performance Comparisons across providers
Security Event Analysis with risk assessment
Capacity Planning with growth projections
Export Capabilities for external analysis

💬 Chat & AI Interaction

🧑‍🔧 Advanced Agent Configuration

Pre-built Agent Types: Coder, Analyst, Creative Writer, Researcher
Custom Agent Creation with specialized prompts and behaviors
Metacognitive Enhancements: Chain of Thought, Visualization of Thought
Voice & Personality Customization for tailored interactions
Memory Systems with episodic and semantic memory support

🗣️ Multimodal Capabilities

Text-to-Speech (TTS) with multiple voice options
Speech Recognition for voice input
Vision Model Support for image analysis and description
Document Processing with OCR and extraction
File Upload & Analysis for various formats

📚 Retrieval Augmented Generation (RAG)

Dynamic Corpus Integration for contextual enhancement
Vector Database Support with ChromaDB integration
Semantic Search with advanced embedding models
Document Chunking with intelligent splitting strategies
Real-time Context Injection for improved responses

⚙️ Advanced Workflows

🧠 Multi-Agent Orchestration

Brainstorm Mode: Collaborative AI teams for ideation
Research Workflows: Multi-source intelligence gathering
Build System: Autonomous software development with specialized agents
Project Management: AI-assisted task planning and execution
Quality Assurance: Automated testing and review processes

📊 Visual Workflow Builder (Nodes)

Drag-and-Drop Interface for workflow design
Pre-built Components: AI models, data processors, connectors
Custom Node Creation with Python scripting support
Conditional Logic with branching and loops
Real-time Execution with progress monitoring

🔗 Integration Capabilities

API Gateway with rate limiting and authentication
Webhook Support for external integrations
Database Connectors for various data sources
File System Integration with secure access controls
Third-party Tool Integration via standardized protocols

🗄️ Knowledge Management

📂 Corpus Management

Text Corpus Creation from various sources
Web Content Integration with intelligent extraction
Document Processing with format support (PDF, DOCX, TXT, MD)
Knowledge Graph Construction for relationship mapping
Version Control for corpus evolution tracking

🕸️ Data Sources

Web Scraping with respect for robots.txt
API Data Integration from external services
Database Connectivity for structured data
File Upload System for corpus ingestion
Real-time Data Feeds for dynamic knowledge

🔍 Search & Discovery

Full-text Search across all knowledge bases
Semantic Search with embedding-based similarity
Faceted Search with filters and categories
Search Analytics for usage optimization
Personalized Recommendations based on user patterns

🛠️ Model Management

📋 Model Operations

Local Model Management: List, download, update, remove Ollama models
Model Information: Detailed specs, capabilities, and performance metrics
Health Monitoring: Model availability and performance tracking
Usage Analytics: Token consumption and cost analysis
Model Comparison: Side-by-side performance evaluation

⚙️ Server Configuration

Ollama Server Management: Start, stop, configure, monitor
Resource Allocation: CPU, memory, and GPU settings
Concurrency Controls: Request queuing and parallel processing
Performance Tuning: Optimization for specific workloads
Backup & Recovery: Model and configuration backup

☁️ External Provider Management

API Key Management: Secure storage and rotation
Rate Limit Monitoring: Usage tracking and alerts
Cost Management: Budget controls and spending alerts
Provider Comparison: Performance and cost analysis
Failover Configuration: Automatic provider switching

📊 Testing & Evaluation

🧪 Model Testing

Feature Testing: JSON handling, function calling, structured output
Performance Benchmarking: Speed, accuracy, and consistency
Comparative Analysis: Multi-model evaluation on same tasks
Regression Testing: Automated testing for model updates
Custom Test Suites: Domain-specific evaluation frameworks

🎯 Quality Assurance

Response Quality Metrics: Coherence, relevance, factuality
Contextual Understanding: Multi-turn conversation evaluation
Bias Detection: Automated bias and fairness testing
Safety Evaluation: Content safety and harmfulness detection
Performance Profiling: Resource usage and optimization

👁️ Vision Model Testing

Image Understanding: Object detection and scene analysis
OCR Capabilities: Text extraction from images
Visual Question Answering: Image-based reasoning
Art and Design: Creative image analysis
Medical Imaging: Specialized vision model testing

🔧 Installation

📋 System Requirements

Python: 3.10+
Operating System: Windows, macOS, Linux
Memory: 8GB+ RAM (16GB+ recommended)
Storage: 10GB+ free space
Network: Internet connection for model downloads

⚡ Automated Setup (Recommended)

# Clone the repository
git clone https://github.com/marc-shade/Ollama-Workbench.git
cd Ollama-Workbench

# Run automated setup
python3 scripts/setup_workbench.py

This script automatically:

✅ Creates virtual environment
✅ Installs all dependencies
✅ Downloads and configures Ollama
✅ Initializes security framework
✅ Sets up directories and configuration
✅ Creates startup scripts
✅ Runs basic tests

🐍 Manual Setup

# Create virtual environment (Python 3.11 recommended; 3.10+ supported)
python3 -m venv venv
source venv/bin/activate  # Unix/Linux/macOS
venv\Scripts\activate     # Windows

# Install dependencies
pip install -r requirements.txt

# Install Ollama (if not already installed)
curl -fsSL https://ollama.ai/install.sh | sh  # Unix/Linux/macOS
# For Windows: Download from https://ollama.ai/download

# Start Ollama server
ollama serve

# Download a default model
ollama pull llama3.2:1b

# Start the workbench
streamlit run main.py

🔧 Configuration

Initial configuration is created automatically, but you can customize:

{
  "OLLAMA_HOST": "http://localhost:11434",
  "WORKBENCH_PORT": 8501,
  "ENABLE_OPENAI_COMPAT": true,
  "ENABLE_ENHANCED_SECURITY": true,
  "ENABLE_AUTH": false,
  "ENABLE_RBAC": true,
  "ENABLE_AUDIT_LOGGING": true,
  "ENABLE_ENCRYPTION": true,
  "ENABLE_OBSERVABILITY": true
}

The UI theme lives in .streamlit/config.toml — colors, fonts, radii, and the sidebar palette are all editable there (Streamlit >= 1.47 native theming).

📚 Documentation

📖 Complete Documentation Suite

Technical Architecture - System design and architecture
Security & Compliance - Security features and compliance
API Documentation - Pipeline framework and APIs
Deployment Guide - Production deployment strategies
Contributing Guide - Development guidelines
User Stories - Detailed use cases and workflows
UX Design Guide - Interface design principles
Observability Specification - Monitoring and tracing
Glossary - Domain terminology and concepts
Apple Silicon Setup - macOS arm64 notes

🤝 Contributing

We welcome contributions from the community! Please see our Contributing Guide for details.

🛠️ Development Setup

# Clone and setup development environment
git clone https://github.com/marc-shade/Ollama-Workbench.git
cd Ollama-Workbench
python3 scripts/setup_workbench.py

# Install development dependencies
pip install -r requirements-dev.txt

# Run the test suite (1,076 tests, hermetic - no live Ollama needed)
python -m pytest tests/

# Run linting
ruff check .
flake8 .

📝 Contributing Areas

🐛 Bug Reports: Help us identify and fix issues
🚀 Feature Requests: Suggest new capabilities
📖 Documentation: Improve guides and tutorials
🧪 Testing: Add test cases and improve coverage
🔒 Security: Enhance security features and audit code
🎨 UI/UX: Improve user interface and experience
🔌 Integrations: Add support for new AI providers

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🆘 Support & Community

📧 Issues: GitHub Issues
💬 Discussions: GitHub Discussions
📚 Documentation: Complete guides in /docs directory
🐛 Bug Reports: Use issue templates for better tracking
💡 Feature Requests: Share your ideas for improvements

🙏 Acknowledgments

Ollama Team - For the excellent local LLM runtime
Streamlit - For the amazing web app framework
OpenAI, Groq, Mistral - For their powerful AI APIs
Opik - For comprehensive LLM observability
Contributors - For making this project better every day

Made with ❤️ by the Ollama Workbench Community

A local-first workbench for serious work with open models.

Name		Name	Last commit message	Last commit date
.claude		.claude
.github		.github
.secure		.secure
.streamlit		.streamlit
assets		assets
autogen_compat		autogen_compat
canvas		canvas
data		data
docs		docs
extension		extension
logs		logs
observability		observability
ollama_workbench		ollama_workbench
persona_lab		persona_lab
prompts		prompts
scripts		scripts
security		security
test_results		test_results
tests		tests
tmp		tmp
tts_server		tts_server
.gitignore		.gitignore
.gitleaks.toml		.gitleaks.toml
.pre-commit-config.yaml		.pre-commit-config.yaml
CLAUDE.md		CLAUDE.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
ROADMAP.md		ROADMAP.md
app_settings.json		app_settings.json
brainstorm_agents_settings.json		brainstorm_agents_settings.json
chat-settings.json		chat-settings.json
cleanup_tts_server.sh		cleanup_tts_server.sh
ensure_ollama_running.sh		ensure_ollama_running.sh
fix_dependencies.sh		fix_dependencies.sh
main.py		main.py
multimodel-chat-settings.json		multimodel-chat-settings.json
native_host_manifest.json		native_host_manifest.json
poetry.lock		poetry.lock
pyproject.toml		pyproject.toml
requirements-dev.txt		requirements-dev.txt
requirements.txt		requirements.txt
ruff.toml		ruff.toml
run_ollama_workbench.sh		run_ollama_workbench.sh
start_ollama_workbench.sh		start_ollama_workbench.sh
start_workbench.sh		start_workbench.sh
voice-settings.json		voice-settings.json
voice_settings.json		voice_settings.json
workbench_config.json		workbench_config.json

Latest commit

History

Folders and files

Repository files navigation