SIM-ONE

SIM-ONE Cognitive Control Protocol (mCP) Server

Project Overview

Welcome to the SIM-ONE mCP Server, a sophisticated, multi-protocol cognitive architecture designed to simulate advanced reasoning, emotional intelligence, and metacognitive governance. This server is the backbone of the SIM-ONE framework, providing a powerful platform for developing and orchestrating autonomous AI agents that can perform complex cognitive tasks.

The server is built on a modular, protocol-based architecture, allowing for dynamic loading and execution of various cognitive functions. From deep reasoning and emotional analysis to advanced entity extraction and self-governance, the mCP Server provides the tools to build truly intelligent systems.

Key Features

• Modular Protocol Architecture: Dynamically load and chain “Cognitive Protocols” to create complex workflows.
• Advanced Reasoning (REP): A rule-based engine for deductive, inductive, and abductive reasoning.
• Emotional Intelligence (ESL): Sophisticated, rule-based emotional analysis to understand user sentiment and intent.
• Entity & Relationship Extraction (MTP): Advanced entity recognition (people, places, organizations) and detection of the relationships between them.
• Cognitive Governance Engine: A powerful metacognitive layer that validates the coherence, quality, and resilience of all cognitive processes.
• Persistent Memory: A memory manager that allows the system to learn and recall information across sessions.
• Secure and Scalable: Built with FastAPI, including features like API key authentication, rate limiting, and a configurable setup for production deployment.

Quick Start Guide

Get the server up and running in 5 minutes.

1. Clone the repository:

git clone [repository-url]
cd SIM-ONE

2. Set up a Python virtual environment:

python3 -m venv venv
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

4. Configure environment variables: Create a .env file in the root directory and add the required variables. See the Configuration Guide for details. A minimal example:

MCP_API_KEY="your-secret-api-key"

5. Run the server:

uvicorn mcp_server.main:app --host 0.0.0.0 --port 8000

The server is now running and accessible at http://localhost:8000.

Architecture Overview

The mCP Server consists of several key components that work in tandem:

• API Gateway (main.py): The public-facing interface of the server, handling incoming requests, authentication, and routing.
• Orchestration Engine: The core component that executes workflows by managing and sequencing Cognitive Protocols.
• Protocol Manager: Responsible for dynamically discovering and loading available protocols (e.g., REP, ESL, MTP).
• Cognitive Protocols: Individual modules that perform specific cognitive tasks. Each protocol is self-contained and exposes a standard interface.
• Cognitive Governance Engine: A metacognitive layer that monitors, validates, and governs the outputs of all protocols to ensure coherence and quality.
• Memory Manager: Interfaces with a Redis database to provide persistent memory and session management.

For a more detailed breakdown, please see the full Architecture Documentation.

Installation

For detailed, step-by-step installation instructions for various platforms and production environments, please refer to our comprehensive Installation Guide.

Configuration

The server is configured entirely through environment variables. For a full list of all required and optional variables, their purpose, and example values, please see the Configuration Guide.

Security & Governance

The mCP Server includes a governance layer and structured audit logging. Enable and tune these with environment flags:

• Governance
  • GOV_ENABLE (default: true): Toggles governance evaluation.
  • GOV_MIN_QUALITY (default: 0.6): Minimum acceptable protocol quality score; lower values are flagged.
  • GOV_REQUIRE_COHERENCE (default: false): If true, incoherence triggers a single retry of the step; persistent incoherence aborts the workflow with an error.
• Rate limiting (per endpoint)
  • RATE_LIMIT_EXECUTE (default: 20/minute)
  • RATE_LIMIT_PROTOCOLS (default: 60/minute)
  • RATE_LIMIT_TEMPLATES (default: 60/minute)
  • RATE_LIMIT_SESSION (default: 30/minute)
  • RATE_LIMIT_METRICS (default: 10/minute)

Audit logs are emitted in JSON format by a dedicated audit logger:

• File: security_events.log (rotated daily; 14 backups)
• Location: server working directory (when running from this repo, see code/security_events.log)
• Event examples: recovery_decision, governance_incoherence_detected, governance_abort, execute_completed
• No sensitive content is logged; governance summaries include aggregate scores and booleans only.

Tip: Responses from /execute include a governance_summary field with quality_scores and is_coherent for quick inspection.

Admin model management (MVLM local)

• List available model aliases and show active alias (admin): GET /admin/models
• Activate a model by alias (admin): POST /admin/models/activate with body { "alias": "main" }
• Configuration via .env:
  • MVLM_MODEL_DIRS=main:models/mvlm_gpt2/mvlm_final,enhanced:/opt/models/next
• ACTIVE_MVLM_MODEL=main

Execution Controls

• Timeouts and concurrency
  • PROTOCOL_TIMEOUT_MS (default: 10000): Max time per protocol step.
  • MAX_PARALLEL_PROTOCOLS (default: 4): Cap for parallel step execution.
  • PROTOCOL_TIMEOUTS_MS: Optional per‑protocol overrides as name:ms pairs, comma‑separated (e.g., ReasoningAndExplanationProtocol:15000,EmotionalStateLayerProtocol:8000).
• Quotas and rate limiting
  • API/IP rate limiting configured via RATE_LIMIT_* envs.
  • Optional per‑API‑key quota: API_KEY_QUOTA_PER_MINUTE (default: 0 disables).
• Metrics
  • JSON metrics: GET /metrics (admin)
  • Prometheus metrics: GET /metrics/prometheus (admin unless METRICS_PUBLIC=true)
  • Prometheus example config: see code/config/prometheus.yml.
  • Grafana example dashboard: code/mcp_server/docs/monitoring/grafana_governance_dashboard.json.

Local Monitoring

Run a local Prometheus + Grafana stack to visualize governance/recovery metrics.

• Start services (single node app + Redis + Prometheus + Grafana):
  • docker-compose -f code/docker-compose.override.yml up
• Endpoints:
  • App: http://localhost:8000
  • Prometheus: http://localhost:9090 (scrapes /metrics/prometheus)
  • Grafana: http://localhost:3000 (default admin password: admin)
• Import dashboard:
  • In Grafana, go to Dashboards → Import
  • Upload code/mcp_server/docs/monitoring/grafana_governance_dashboard.json
  • Select the Prometheus data source and save
• Notes:
  • The app exposes Prometheus metrics at /metrics/prometheus; set METRICS_PUBLIC=true for local testing (already set in override compose).
  • Governance counters include coherence failures, governance aborts, recovery retries, and fallbacks.

Model Test CLI

Quickly test a local GPT‑2–style MVLM without running the server.

• Script: code/tools/mvlm_textgen.py
• Requirements: included in code/requirements.txt (transformers, torch, safetensors).
• Usage examples:
  • Project MVLMEngine (simple defaults)
    • python code/tools/mvlm_textgen.py --model-dir models/mvlm_gpt2/mvlm_final --prompt "Give three bullet points on SIM-ONE governance:"
  • Raw Transformers (deterministic greedy)
    • python code/tools/mvlm_textgen.py --engine hf --greedy --max-new-tokens 96 --model-dir models/mvlm_gpt2/mvlm_final --prompt "Write a concise executive summary about SIM-ONE governance:"
  • From a file
    • python code/tools/mvlm_textgen.py --model-dir models/mvlm_gpt2/mvlm_final --prompt-file prompt.txt
  • From stdin
    • echo "Test prompt" | python code/tools/mvlm_textgen.py --model-dir models/mvlm_gpt2/mvlm_final

Notes

• Force CPU if needed: set CUDA_VISIBLE_DEVICES="" or PYTORCH_FORCE_CPU=1.
• The model directory must contain standard HF files (config.json, model.safetensors, tokenizer files).

API Documentation

The server exposes a powerful API for executing cognitive workflows. For detailed information on all available endpoints, request/response formats, authentication, and usage examples, please refer to our full API Documentation.

Basic API Usage Example

Here is a quick example of how to execute a workflow using curl:

curl -X POST "http://localhost:8000/v1/execute" \
-H "Authorization: Bearer your-secret-api-key" \
-H "Content-Type: application/json" \
-d '{
    "workflow": "StandardReasoningWorkflow",
    "data": {
        "user_input": "John works at Microsoft and lives in Seattle."
    }
}'

Contributing

We welcome contributions from the community! If you’d like to contribute, please read our Contributing Guidelines to get started.

License

This project is licensed under the terms of the AGPL v3 / Commercial.

This site is open source. Improve this page.