Aragora -- The Decision Integrity Platform

Version 2.6.3 | Commercial Overview Status: Internal snapshot; metrics are directional unless sourced in docs/STATUS.md.

Executive Summary

Aragora is the Decision Integrity Platform -- orchestrating 30+ agent types to adversarially vet decisions against your organization's knowledge, then delivering audit-ready decision receipts to any channel.

You don't just get an answer. You get a defensible decision trail.

Unlike chatbots, Aragora builds institutional memory with full audit trails. Vetted decisionmaking is the engine. The product is a defensible decision record.

Five Pillars

Aragora is built on five architectural commitments that together produce something no single-model tool can offer.

Pillar	What It Means
1. SMB-Ready, Enterprise-Grade	Useful to a 5-person startup on day one; scales to regulated enterprise without rearchitecting. Security and compliance built in, not bolted on.
2. Leading-Edge Memory and Context	4-tier Continuum Memory, Knowledge Mound (28 registered adapters), and RLM context compression enable coherence across long multi-round sessions and large document sets.
3. Extensible and Modular	Connectors, SDKs (Python + TypeScript, 140 namespaces), 2,000+ API operations, OpenClaw integration, workflow engine, marketplace.
4. Multi-Agent Robustness	Heterogeneous agents (Claude, GPT, Gemini, Grok, Mistral, DeepSeek, Qwen, Kimi) produce outputs more robust, less biased, and higher quality than single models.
5. Self-Healing and Self-Extending	Nomic Loop autonomous improvement, red-team stress-testing, multi-agent code editing with human approval gates.

What Aragora Does

Capability	Description	Business Value
Omnivorous Input	Ingest from documents, APIs, databases, web, voice	Single platform for all information sources
Multi-Channel Access	Query via web, Slack, Telegram, WhatsApp, API	Meet users where they already work
Multi-Agent Consensus	30+ heterogeneous agent types debate to conclusions	Diverse perspectives, defensible decisions
Bidirectional Dialogue	Ask follow-ups, refine questions, drill into details	Interactive human-AI collaboration
Evidence Trails	Cryptographic audit chains with provenance tracking	Compliance-ready documentation
Learn and Improve	4-tier memory with cross-session pattern learning	Continuously improving accuracy

Core Value Proposition

For Engineering Leaders

Reduce review bottlenecks: AI agents provide first-pass critique 24/7
Catch blind spots: Different AI models notice different issues
Accelerate decisions: Get multi-perspective analysis in minutes, not days

For Compliance Officers

Audit trails: Every debate produces a cryptographic receipt
Dissent tracking: Know what was contested and why
Provenance chains: Trace every claim to its source

For Security Teams

Adversarial testing: Built-in red team mode attacks your specs
Gauntlet mode: Systematic stress-testing with risk heatmaps
Pattern learning: System remembers past vulnerabilities

Platform Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                         ARAGORA PLATFORM                                 │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                          │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │                    AGENT LAYER (30+ Agent Types)                  │   │
│  │  Claude │ GPT │ Gemini │ Grok │ Mistral │ DeepSeek │ Qwen │ Kimi │   │
│  │                     + Local Models (Ollama, LM Studio)            │   │
│  └──────────────────────────────┬───────────────────────────────────┘   │
│                                  ▼                                       │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │                    DEBATE ENGINE                                  │   │
│  │  • 9-round structured protocol (Propose → Critique → Synthesize)  │   │
│  │  • Graph debates with branching │ Matrix debates for scenarios    │   │
│  │  • Consensus detection │ Convergence analysis │ Forking support   │   │
│  └──────────────────────────────┬───────────────────────────────────┘   │
│                                  ▼                                       │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │                    KNOWLEDGE LAYER                                │   │
│  │  • Belief networks with Bayesian propagation                      │   │
│  │  • Claims kernel with typed relationships                         │   │
│  │  • Evidence provenance with hash chains                           │   │
│  │  • Citation tracking and reliability scoring                      │   │
│  └──────────────────────────────┬───────────────────────────────────┘   │
│                                  ▼                                       │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │                    MEMORY SYSTEM (4 tiers)                        │   │
│  │  Fast → Medium → Slow → Glacial                                   │   │
│  │  • Surprise-based learning │ Consolidation scoring                │   │
│  │  • Cross-session pattern extraction                               │   │
│  └──────────────────────────────┬───────────────────────────────────┘   │
│                                  ▼                                       │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │                    OUTPUT LAYER                                   │   │
│  │  Decision Receipts │ Risk Heatmaps │ Dissent Trails │ Proofs     │   │
│  └──────────────────────────────────────────────────────────────────┘   │
│                                                                          │
└─────────────────────────────────────────────────────────────────────────┘

Commercial Readiness Assessment

Overall: 90% Production Ready (internal estimate)

Category	Score	Status	Notes
Error Handling & Resilience	95%	Ready	Circuit breakers, retry policies, graceful degradation
Security & Authentication	92%	Ready	OIDC/SAML, MFA, encryption, audit logging
Scalability & Performance	92%	Ready	Connection pooling, caching, rate limiting
Observability & Monitoring	90%	Ready	Prometheus, Grafana, OpenTelemetry
Testing & QA	95%	Ready	130,000+ tests across 3,000+ files
Documentation	91%	Ready	API docs, runbooks, compliance guides
Compliance & Governance	88%	Ready	RBAC v2 with 50+ permissions, role hierarchy
SDK & Integrations	90%	Ready	140 TypeScript namespaces, 8 bot handlers
OVERALL	90%	SMB Ready	Enterprise-grade features integrated

Deployment Readiness

Docker container with non-root user
Kubernetes manifests in /deploy/k8s/
Health checks and readiness probes
Prometheus metrics endpoint
Grafana dashboards included

Key Differentiators

1. Heterogeneous Agent Orchestration

Unlike single-model solutions, Aragora runs debates across 30+ agent types/providers. Different models catch different issues—Claude excels at reasoning, GPT at breadth, Gemini at design, Grok at lateral thinking.

2. Audit-Ready Output

Every debate produces a Decision Receipt with:

Cryptographic hash chain
Evidence provenance
Dissent tracking
Timestamp verification

3. Adversarial Testing Built-In

Gauntlet Mode provides systematic stress-testing:

Security red-team attacks
Devil's advocate logic testing
Scaling critic analysis
Compliance verification (GDPR, HIPAA, SOC 2, AI Act)

4. Learning Memory System

The 4-tier Continuum Memory enables:

Pattern extraction from successful critiques
Cross-session learning
Institutional knowledge accumulation
Surprise-based prioritization

5. Enterprise-Grade Security

OIDC/SAML SSO integration
MFA support (TOTP/HOTP)
AES-256-GCM encryption at rest
Multi-tenant isolation with quotas

Use Cases

Specification Review

aragora gauntlet spec.md --profile thorough --output receipt.html

Stress-test API specifications, architecture documents, and technical designs before implementation.

Compliance Audit

aragora gauntlet policy.yaml --input-type policy --persona gdpr

Automated compliance checking against GDPR, HIPAA, SOC 2, and AI Act requirements.

Code Review

git diff main | aragora review

AI red-team review of pull requests with unanimous consensus highlighting.

Decision Validation

from aragora import Arena, Environment, DebateProtocol

env = Environment(task="Should we adopt microservices?")
protocol = DebateProtocol(rounds=5, consensus="majority")
arena = Arena(env, agents, protocol)
result = await arena.run()

Structured debate for strategic decisions with evidence-based recommendations.

Platform Statistics

Metric	Source
Codebase size & tests	See `docs/STATUS.md`
Agent catalog	`AGENTS.md`
Connector catalog	`docs/CONNECTORS.md`
Memory tiers	4 (see `docs/MEMORY_TIERS.md`)

Deployment Options

Self-Hosted

Docker Compose for single-node deployment
Kubernetes for scale-out deployment
Supports SQLite (dev) or PostgreSQL (prod)

Cloud

AWS Lightsail (current production)
Any Kubernetes-compatible cloud (AWS EKS, GCP GKE, Azure AKS)
Cloudflare Tunnel for secure ingress

Hybrid

On-premises control plane
Cloud-based agent APIs
Air-gapped deployment support

Integration Points

Chat Platforms

Slack (bot + connector)
Discord (bot + connector)
Microsoft Teams (bot + connector)
Google Chat (connector)

Data Sources

GitHub, GitLab
SharePoint, Confluence, Notion
ArXiv, Wikipedia, news APIs
Healthcare systems (HL7/FHIR)
SEC filings, legal databases

Observability

Prometheus metrics export
Grafana dashboards included
OpenTelemetry tracing
SIEM integration

Enterprise Readiness

Capability	Resolution	Status
Fine-grained RBAC	RBAC v2 with 7 roles, 50+ permissions	Complete
Automated backups	BackupManager with incremental support	Complete
Bot handler consolidation	BotHandlerMixin across 8 platforms	Complete
TypeScript SDK	140 namespaces wired to client	Complete
OpenClaw integration	Portable agent governance	Complete
Knowledge Mound Phase A2	Contradiction detection, confidence decay, RBAC governance	Complete
SLA documentation	Legally-binding service levels	In Progress
Distributed rate limiting	Redis-backed cluster-aware limiting	In Progress

Pricing Considerations

Placeholder for commercial discussion

Potential Models

Per-seat licensing - Based on user count
Usage-based - Per debate/API call
Tier-based - SMB / Enterprise / Enterprise+
Hybrid - Base license + usage overage

Cost Factors

AI provider API costs (passed through or absorbed)
Compute and storage
Support tier (community, business, enterprise)
Compliance certifications

Getting Started

Quick Start (5 minutes)

git clone https://github.com/an0mium/aragora.git
cd aragora
pip install -e .
export ANTHROPIC_API_KEY=your-key
aragora ask "Design a rate limiter" --agents anthropic-api,openai-api

Production Deployment

See PRODUCTION_READINESS.md for the complete checklist.

API Integration

See SDK_GUIDE.md for the Python SDK reference.

Contact

Domain: aragora.ai
Documentation: docs/
API Reference: API_REFERENCE.md

Document generated from comprehensive codebase exploration. Feature counts verified against actual module inventory (February 2026).

Executive Summary​

Five Pillars​

What Aragora Does​

Core Value Proposition​

For Engineering Leaders​

For Compliance Officers​

For Security Teams​

Platform Architecture​

Commercial Readiness Assessment​

Overall: 90% Production Ready (internal estimate)​

Deployment Readiness​

Key Differentiators​

1. Heterogeneous Agent Orchestration​

2. Audit-Ready Output​

3. Adversarial Testing Built-In​

4. Learning Memory System​

5. Enterprise-Grade Security​

Use Cases​

Specification Review​

Compliance Audit​

Code Review​

Decision Validation​

Platform Statistics​

Deployment Options​

Self-Hosted​

Cloud​

Hybrid​

Integration Points​

Chat Platforms​

Data Sources​

Observability​

Enterprise Readiness​

Pricing Considerations​

Potential Models​

Cost Factors​

Getting Started​

Quick Start (5 minutes)​

Production Deployment​

API Integration​

Contact​