AI infrastructure for builders

The work nobody has time for — handled.

ElfTech builds AI systems that own the operational work surrounding software and business — the product management, the deployments, the marketing, the field reporting, the compliance documentation — so you can focus on what matters.

01
Building has never been easier.
AI tools have collapsed the cost of creating software, content, and campaigns. Every team is building more, faster, across more domains.
02
Managing has never been harder.
Every tool without a product owner, every campaign without analytics, every field report that never reaches a compliance record — the operational gap is growing faster than headcount.
03
AI should close the gap it created.
Not as an assistant. As an owner. ElfTech builds autonomous AI systems — with defined boundaries and human oversight — that own operational work end to end.
Infrastructure for anyone building software.
Horizontal tools that serve any organization with internal software — regardless of industry. They solve structural problems that exist everywhere.
Steward In Development
The trust interface between human and autonomous system. Triage flows through Armature, high-stakes decisions through Quorum, code changes through AnvilWorks. Every escalation is earned.
Watches repos and classifies all incoming work
Routes triage to Armature, decisions to Quorum, code to AnvilWorks
Autonomous execution within defined boundaries
Built on the full ElfTech platform stack
Learn more
steward / portfolio
DeployHub1 approval8712m
Config Serviceattention722h
Status CLI911d
Auth Gateway1 approval834h
Greenroom In Development
On-premise frontend deployment for teams who work with data that can't leave the building. The Netlify experience — git-push deploys, preview URLs, instant rollbacks — entirely inside your firewall.
Push to GitHub, get a live URL in under 60 seconds
HIPAA, FedRAMP, SOC 2 — no data leaves your network
Bundled HTTPS and DNS — no internet required
Ships as a single docker compose up
Learn more
greenroom / deployments
patient-portalmainportal.internal.co2m
trading-dashfeat/chartspr-47.trading.internal.co8m
design-systemmainbuilding…now
claims-reviewmainclaims.internal.co1h
The trust layers that make everything else possible.
Two capabilities that power every ElfTech product — and are independently valuable as services for organizations building their own AI infrastructure.
Quorum
The Reasoning Layer

Structured multi-agent deliberation for decisions that matter. Teams of AI agents with distinct expert perspectives argue, challenge, and synthesize through formal debate protocols — producing auditable reasoning records, not black-box answers.

Three-round invariant structure — predictable cost
`dissenting_opinions` field — minority views preserved
DocGen: executive brief, ADR, risk report artifacts
Academic lineage: DeepMind Co-Scientist research
Explore Quorum
deliberation / LDAP authconf: 0.87
◻ Facts● Instinct● Risks● Value● Alternatives● Synthesis
⚗️
Alembic
The Intelligence Layer

Domain-expert small language models, fine-tuned not just for knowledge but for capability — then deployed on your infrastructure where your data never leaves. Coupled with Quorum for reasoning quality that cloud APIs can't match on your domain.

LoRA fine-tuning for domain-specific capabilities
On-premise deployment — data never leaves your walls
75%+ cost reduction vs. cloud LLM APIs
Integrated with Quorum for auditable reasoning
Explore Alembic
alembic / deployed models3 active
legal-contracts-7bQwen 3.5 + LoRA94%
clinical-pathways-3bPhi-4 + LoRA91%
construction-safety-7bMistral + LoRA89%
Armature
The Workflow Harness Layer

A YAML-first multi-agent workflow harness — open source and MIT licensed. Define researcher, worker, and judge agents as a DAG. Execute in parallel. Then let the system study its own traces and improve automatically: every run, every time. A maturity of agents that gets smarter as it works.

YAML spec → validated DAG → parallel execution
Self-improvement via Implicit Harness Rating (IHR) and SpecRefiner
`mission:`, `continuation:`, `triggers:` — long-horizon workflow support
LiteLLM provider layer — any model, any cloud or local
Open Source · MIT
Explore Armature
armature / runIHR=0.91
DAG validated (3 stages)
researcher running...
researcher done (1.4s)
analyst running...
analyst done (2.2s)
editor done (0.9s, score=8.7)
traces/run-20260517.json
AnvilWorks
The Execution Layer

A stateful coding execution engine — CLI-first, scriptable, and chainable. Give it a natural-language goal and a project path; it runs the AI coding agent, validates with tests, and commits to git. Use it at the terminal, drop it in a script, or chain it from another system.

anvilworks run for quick jobs · anvilworks project for full builds
Single-job mode (minutes) to multi-phase project execution (days to weeks)
Clarify → implement → validate → quality gate → repair loop
Autonomous failure recovery — human escalation only when genuinely stuck
Learn more
anvilworks / terminal● running
$ anvilworks run -p /srv/auth \
  "Add rate limiting to login endpoint"
Clarifying acceptance criteria...
Implementing [████████░░░] 78%
Tests passed (14/14)
Committed: a4f7c9d · feat/rate-limit
AI that owns the work in specific domains.
Vertical solutions built on ELF's agentic platform and deep-reasoning RAG engine. Each one targets a domain where the operational gap is acute, the data is complex, and the existing tools aren't cutting it.
📣
Herald Early Stage
AI marketing teams for local businesses. Content creation, social media, campaign optimization, and analytics — run by coordinated AI agent teams, not a single chatbot.
7 specialized AI teams (content, social, analytics, brand…)
POS integration for data-driven campaigns
Multi-platform publishing (Instagram, TikTok, email, SMS)
Brand compliance built into every output
Market → Restaurants, retail, local services
🩺
Reconnect Early Stage
AI-powered patient reactivation for elective medical practices. Analyzes dormant patient lists, identifies reactivation opportunities, and runs targeted campaigns — HIPAA compliant from the ground up.
HIPAA-compliant patient data analysis
Dormant patient scoring and segmentation
Multi-channel reactivation campaigns
15–25% reactivation rate target
Market → Med spas, cosmetic surgery, dermatology
🔷
Tessera Available
Document intelligence that discovers what you didn't know to look for. Analyzes your entire corpus, surfaces emergent patterns, and projects a living thematic map across every document — without being told what matters.
Emergent schema discovery — no predefined taxonomy required
Knowledge graph + vector hybrid retrieval
Cross-document synthesis with evidence citation
Configurable for legal, compliance, and enterprise research
Market → Legal discovery, compliance, financial analysis
🏗️
Relay Planned
AI-powered field intelligence for industrial construction. Captures reports through Telegram — annotated photos, voice memos, text — and handles the classification, compliance routing, and documentation that large projects demand.
Field capture via Telegram — photos, annotations, voice, text
AI classification, routing, and compliance documentation
Multi-site portfolio view across active projects
Audit-grade trails for multinational client reporting
Market → Industrial GCs, oil & gas, mining, energy, heavy construction

Built on ELF

Every ElfTech product runs on ELF — a purpose-built AI orchestration platform with multi-agent teams, MCP servers, and Kubernetes-native deployment. ELF isn't a product you buy. It's the engine underneath everything we ship.

When we say "AI that owns the work," we mean coordinated teams of AI agents with defined roles, debate frameworks for decision quality, and human oversight at the boundaries. Not a single model with a prompt.

700K+
Lines of orchestration code
Multi-Agent
Coordinated AI teams, not single models
7-Stage
Deep-reasoning RAG pipeline
K8s Native
Production-grade infrastructure

Why ELF

Every organization has work that's essential but unglamorous. The maintenance, the triage, the follow-ups, the campaigns, the field reports, the compliance documentation. The things that keep the lights on but don't make the roadmap.

ElfTech builds AI systems that own this work outright. Not copilots. Not assistants. Autonomous systems with defined boundaries, transparent reasoning, and human oversight where it matters.

E
Effective
AI that delivers outcomes, not suggestions
L
Labor
Real work — not demos, not prototypes
F
Force
Teams of AI agents, not individual tools