title	Inference
emoji	🌌
colorFrom	indigo
colorTo	purple
sdk	docker
app_port	7860

Inference

Independent Multi-Agent Intelligence System
Master Architecture, Multi-Model Reasoning, Debate Engine, and FRIDAY Integration.

---

🌟 Overview

Inference is a local-first, provider-agnostic multi-agent intelligence platform. Its purpose is to make multiple AI systems work together so that you can ask one complex question and receive a stronger, critically evaluated, and empirically verified answer than any single model would provide alone.

Core Principles

• Inference is not FRIDAY, and FRIDAY is not a wrapper around Inference. They are two independent intelligent systems. Later, Inference can exchange specialized reasoning, analysis, and debate outcomes with FRIDAY through a secure API boundary.
• Multi-Model / Provider Agnostic: Connects to Gemini, Groq, Mistral, OpenRouter, NVIDIA, Cohere, and HuggingFace through a unified provider gateway.
• Structured Debate Protocol: Uses multi-agent adversarial critique, rebuttal, evidence checking, and uncertainty-aware synthesis.
• Persistent Memory & Learning: Persistent SQLite memory with strict agent scoping, evaluation rubrics, and routing strategy optimization.
• Local-First, Cloud Inference: Orchestration, storage, and UI run locally on your laptop, while model inference is dispatched to fast cloud provider APIs without GPU strain.

---

🏛️ System Architecture

                      USER / CLIENT
                            |
                            v
                 +----------------------+
                 |     FastAPI / UI     |
                 +----------+-----------+
                            |
                            v
                 +----------------------+
                 |  ORCHESTRATOR CORE   |
                 +----------+-----------+
                            |
         +------------------+------------------+
         |                  |                  |
         v                  v                  v
    Task Router       Agent Registry        Memory
         |                  |                  |
         +------------------+------------------+
                            |
                            v
                 +----------------------+
                 |    Debate Engine     |
                 | (10 Specialist Roles)|
                 +----------+-----------+
                            |
                            v
                 +----------------------+
                 | Evaluator / FactCheck|
                 +----------+-----------+
                            |
                            v
                 +----------------------+
                 |     Synthesizer      |
                 | (Audit Trail + Conf) |
                 +----------+-----------+
                            |
                            v
                 +----------------------+
                 |  Learning & Strategy |
                 +----------------------+

---

🤖 Initial 10-Specialist Agent Panel

Agent Role	Primary Responsibility	Typical Use Case
Researcher	Find & organize relevant information	Knowledge synthesis, comparisons
Architect	Design systems and structure	Software design, technical planning
Coder	Propose implementation approaches	Code generation, refactoring
Debugger	Trace failures & inconsistencies	Root-cause analysis, error diagnosis
Security Analyst	Identify vulnerabilities & risks	Threat modeling, safe design
Data Analyst	Reason from structured data	Metrics, quantitative analysis
Critic	Attack weak assumptions & fallacies	Red-team reasoning, counterexamples
Fact Checker	Separate claims from evidence	Claim verification, contradiction detection
Strategist	Compare alternatives & tradeoffs	Prioritization, decision support
Synthesizer	Produce final coherent answer	Integrates claims, reports uncertainty
Trading Analyst	Analyze quantitative trading metrics & drawdown	Strategy parameter advice (SL/TP, leverage)

📈 Algorithmic Trading Bot Integration (Advisory Authority)

• Specialist Agent (TradingAnalyst): Evaluates performance telemetry (win rate, profit factor, drawdown, consecutive loss streaks) sent by FRIDAY.
• Debate & Consultation: Engages in multi-agent deliberation with the Strategist and Critic to formulate calibrated recommendations (e.g. tightening stop loss, adjusting position sizes).
• Strict Safety Boundary: Inference NEVER calls exchange APIs or executes trades directly; it returns structured AIUniverseDecision recommendations for FRIDAY to record and present to the user.

---

📖 Project Diary & Memory Standard

Development in this repository strictly adheres to the project memory and diary policy:

• Master Index: inference_DIARY.md
• Specification Standard: inference_DIARY_SPEC.md
• Daily Logs: Located in diary/ (e.g. diary/2026-08-24.md)

---

🚀 Getting Started

1. Clone & Setup

git clone https://github.com/surendra2304/Inference.git
cd Inference
python -m venv .venv
# On Windows:
.venv\Scripts\activate
# On Linux/macOS:
source .venv/bin/activate
pip install -e ".[dev]"

2. Configure Environment Variables

cp .env.example .env
# Edit .env and supply your API keys (GEMINI_API_KEY, GROQ_API_KEY, etc.)

To inspect configured provider-key counts locally without printing credential values, run python -m app.core.key_inventory from this repository. The report counts deduplicated values recognized by the current settings model. It does not verify that credentials work, belong to separate provider accounts, or have available quota; those require separate provider checks. Do not hardcode the count in ecosystem manifests because the local and deployed environment pools can change independently.

3. Run the Service

uvicorn app.main:app --reload --port 8000

4. Or Run It With Docker Compose

# Optional but recommended: supply provider keys
cp .env.example .env

docker compose -f docker-compose.production.yml up --build
# API on http://localhost:8000  (health: /health, dashboard: /ui)

The compose file provisions a single API container with a persistent volume at /app/data for the SQLite store. The .env file is optional at start-up: the container boots and serves /health without it, but model calls fail closed with 503 until provider keys are configured.

---

📚 Documentation

• Architecture & Design
• Specialist Agents
• REST API Specifications
• Trading Consultation API
• Memory & Persistence
• Experiments & Benchmarks