# Podesta - AI Infrastructure Built on Rigor

> The bureau of verified intelligence. Hephaistos orchestrates. Prometheus
> connects. Mythos converses. Lumen verifies. Four systems, one standard:
> every output is accountable.

Contact: hello@podesta.ai
Canonical URL: https://podesta.ai

## The Exhibits

### Exhibit I: Workflows Break in Silence
Pipelines fail without notice. Data rots in queues. Nobody knows until a customer calls.

### Exhibit II: AI Outputs Without Accountability
Models hallucinate. Summaries fabricate. There is no paper trail for what was generated or why.

### Exhibit III: Retrieval Without Rigor
RAG systems return noise dressed as signal. No verification, no source attribution, no confidence measure.

## The Four Departments

### Hephaistos - Workflow Orchestration

Define AI workflows in a type-safe language, let Hephaistos execute them. Automatic parallelization, multi-provider orchestration, full audit trails.

Features:
- **Automatic parallelization** - Independent tasks run concurrently by default
- **Multi-provider** - Mix OpenAI, Anthropic, Google in one workflow
- **Human-in-the-loop** - Checkpoint gates for manual review


**DSL - Akribes:** Type-safe .akr workflow definitions compiled to execution graphs. Full IDE support via LSP.

Department reference: DEPT. ORCHESTRATION / HEPHAISTOS-SYS

### Prometheus - Platform Protocol Layer

The lingua franca of the platform. A protocol layer that lets Hephaistos, Mythos, and Lumen communicate without translation overhead.

Features:
- **Unified protocol** - One vocabulary for inter-service communication
- **Real-time messaging** - Bidirectional with automatic reconnection
- **Correlation tracking** - Every request threaded across all services


Department reference: DEPT. INFRASTRUCTURE / PROMETHEUS-SYS

### Mythos - Conversational AI Processing

Multi-turn conversational AI with document-aware processing. Every answer backed by evidence, not hope.

Features:
- **Document-aware** - Upload and reason over PDFs, images, text with OCR
- **Integrated tools** - Document, workflow, retrieval, and verification tools
- **Full audit trail** - Every tool call, token count, and verification logged

**Lumen-Verified Conversations:** Every response cross-referenced against your knowledge base in real time. Claims verified, sources cited, hallucinations flagged — all mid-conversation.


Department reference: DEPT. INTELLIGENCE / MYTHOS-SYS

### Lumen - Verification & Trusted Retrieval

Every AI output is verifiable. Cross-references claims against source material, scores confidence, provides attribution chains. The trust layer for everything else.

Features:
- **Claim verification** - Decompose output, check against evidence, score confidence
- **Attribution chains** - Every claim traced to origin via NLI models
- **Adaptive depth** - Fast, standard, or full verification based on confidence


**Enrichment - Context-Engine:** 4-stage CRAG pipeline — query decomposition, concurrent retrieval, deduplication, token-budget assembly.

Department reference: DEPT. VERIFICATION / LUMEN-SYS

## Division of Responsibilities

The bureau handles: service uptime and availability; infrastructure provisioning and scaling; security patching and compliance; model provider management; monitoring, alerting, and incident response; data encryption at rest and in transit. Uptime is Podesta's responsibility — not yours.

You focus on: defining workflows and business logic; uploading your knowledge base; reviewing verification results; integrating via API; deciding what to ship.

## Built on Principles

- **Verifiable by Default** - Every AI output links to its source.
- **Full Audit Trails** - Every decision, tool call, and token tracked.
- **Type-Safe Pipelines** - Akribes workflows are compiled, not scripted.
- **Provider Agnostic** - Swap models without rewriting workflows.
- **Isolated by Design** - Tenant data never crosses boundaries.
- **Real-Time, Not Batch** - Streaming responses, live verification.