Lattice
LatticeAgent HQ
AGENT HEADQUARTERS

The Open-Source
Headquarters for AI Agents

Lattice is where AI agents get their identity, their permissions, their compute, and their orders. From a weekend hack to Fortune 500 governance — self-hosted, vendor-neutral, runs anywhere.

5
Components
280K+
Lines of Code
6
Deploy Targets
Apache 2.0
Open Source
brew install latticehq/lattice/lattice
Get StartedView on GitHub
lattice — agent_hq — zsh — 80×24
zshlattice v0.1.0utf-8
THE CHALLENGE

Enterprise Trust Requires Scale

Historically, meeting enterprise security requirements has required large organizations with dedicated platform teams.

Application-Level Enforcement

Every AI company re-implements SSO, RBAC, audit logs, and compliance controls inside their application code.

  • ✗Expensive engineering resources
  • ✗Diverts from core product development
  • ✗Trust depends on code discipline

Coupled to Team Size

Enterprise readiness becomes tightly coupled to organizational scale and team maturity.

  • ✗Small teams can't compete
  • ✗Innovation slowed by infrastructure work
  • ✗Enterprise deals require large teams

The result: Only large organizations can ship software that enterprises trust.

This creates an artificial barrier between individual innovation and enterprise deployment.

THE SOLUTION

Move Enforcement Into the Runtime

Lattice decouples trust from team size by providing a complete headquarters — enforcement, compute, operations, and community — as open-source infrastructure.

1

Declare Constraints

Agent developers declare identity, authorization, audit, and deployment constraints once.

2

Runtime Enforcement

Lattice sits in the execution path and enforces constraints before actions are executed.

3

Violations Blocked

Policy violations are structurally impossible — enforcement happens by design, not by discipline.

For Enterprises

  • Enforced by Runtime
    Trust is structural, not based on vendor promises
  • Auditable & Provable
    All enforcement decisions are transparent and inspectable
  • Deploy Anywhere
    Cloud, self-hosted, or air-gapped environments

For Developers

  • Build Capabilities, Not Infrastructure
    Focus on agents, not enforcement systems
  • Enterprise-Ready from Day One
    Ship to enterprises without a platform team
  • Zero API Costs
    Local inference on your hardware, no monthly bill

From solo hacker to Fortune 500 — same platform, same codebase, same community.

Your agents. Your models. Your rules. Your infrastructure.

Lattice: The Open-Source Trust Layer for the AI Agent Era
THE PLATFORM

Agent Headquarters

Five integrated components. One open-source platform. Everything your agents need.

Runtime

The Kernel

Enforcement engine — identity, authorization, audit, deployment constraints. Every agent action passes through four gates.

Go · Apache 2.0

Inference

The Compute

Local LLM serving on your hardware. MLX for Apple Silicon, CUDA for NVIDIA. mDNS autodiscovery forms GPU clusters automatically.

Go + Python · Apache 2.0

Workbench

The Interface

Agent IDE and operations console. Multi-model chat, real-time monitoring, cost tracking. Desktop, web, CLI, VS Code.

TypeScript · MIT

Registry

The Ecosystem

Community templates, modules, and presets for Docker, Kubernetes, AWS, GCP, Azure. Deploy agents anywhere in minutes.

Terraform · Apache 2.0

Enterprise

The Scale Layer

Admin control planes, policy lifecycle, directory integration (AD, Okta, LDAP), compliance exports (SOC2, HIPAA, FedRAMP).

Commercial · Self-Hosted
Lattice Agent Headquarters — the complete open-source platform for AI agents
ENFORCEMENT PRIMITIVES

What Lattice Enforces

Four core enforcement layers that operate in the execution path, making violations impossible by construction.

Identity

Who or what is making the request?

Cryptographically verified identity for every agent, user, and service across cloud, self-hosted, and air-gapped environments.

  • »OAuth 2.0, OIDC, SAML, mutual TLS, API keys
  • »Unique cryptographic principal per agent
  • »Cross-environment identity federation
  • »Zero-trust identity model

Authorization

Is this action allowed?

Policy-as-code evaluated at runtime, not coded into applications. Fine-grained control over what agents can do.

  • »RBAC + attribute-based policies (ABAC)
  • »Fine-grained resource permissions
  • »Dynamic real-time policy evaluation
  • »Cross-platform consistency

Audit

What happened and when?

Immutable, tamper-evident records of every enforcement decision and agent action for compliance and forensics.

  • »Cryptographically chained audit trail
  • »Compliance reporting (SOC2, HIPAA, FedRAMP, GDPR)
  • »Queryable event history
  • »Real-time event streaming

Deployment Constraints

Where can this run?

Structural boundaries on agent execution — where they can run, what resources they consume, and when.

  • »Environment and geographic restrictions
  • »Network segmentation enforcement
  • »Resource quota management
  • »Air-gap validation
LOCAL INFERENCE

Your Models. Your Hardware. Zero API Costs.

Run LLMs on your own machines with an OpenAI-compatible API. Data never leaves your infrastructure.

M4
Apple Silicon (MLX)
Optimized for M1-M4 Metal GPU. Native performance on Mac.
NVIDIA (CUDA)
llama.cpp with CUDA. Multi-GPU automatic tensor splitting.
OpenAI-Compatible API
/v1/chat/completions — any tool that speaks OpenAI works with Lattice.
Zero-Config GPU Clustering
mDNS autodiscovery. Machines on the same network form a cluster automatically.
Multi-Model Pool
LRU eviction, memory budgeting. Serve multiple models simultaneously.
Solo Developer
# One command — local inference running
$ lattice inference serve \
    --model mlx-community/Llama-3.2-3B-Instruct-4bit
✓ OpenAI-compatible API at localhost:8000
Team GPU Cluster
# Machine 1 — starts cluster automatically
$ lattice inference serve \
    --model meta-llama/Llama-3.1-70B --cluster
# Machine 2, 3, 4 — discover and join via mDNS
$ lattice inference serve --join
✓ Model distributed across 4 nodes. Zero config.
MULTI-CLOUD GOVERNANCE

The Governance Layer Every Cloud Needs

Unified governance across every agent platform, every cloud, every environment. Vendor-neutral by design.

Vertex AI Agents
Bedrock Agents
Azure AI Agents
Agentforce
Custom Agents
OSS Agents
LATTICE
One identity system
One auth engine
One audit trail
One policy framework
Self-hosted. Auditable. Vendor-neutral. Yours.

Multi-Cloud Enterprises

Bank deploying across GCP + on-prem? Hospital spanning multiple systems? Lattice provides unified identity, auth, and audit across every environment.

Air-Gapped Deployments

Defense, healthcare, government — Lattice runs fully offline with zero internet access. Complete governance in disconnected environments.

Lattice: The Open-Source Infrastructure for AI Agent Governance
FULL STACK ARCHITECTURE

How Lattice Works

Five components working together — from enforcement kernel to developer interface.

YOUR AGENTS · LangChain · CrewAI · AutoGen · Custom
L A T T I C E   A G E N T   H Q
WORKBENCH
Agent Chat · Operations · Monitoring · Desktop / Web / CLI / VS Code
RUNTIME
Identity
Auth
Audit
Constraints
INFERENCE
MLX · CUDA · OpenAI API
REGISTRY
Templates · Modules · Presets
YOUR INFRASTRUCTURE · Laptop · Homelab · Docker · K8s · Cloud · Air Gap

Open Source (Apache 2.0 / MIT)

Runtime, Inference, Workbench, and Registry are fully open source and auditable.

  • ✓Identity and authorization evaluation
  • ✓Policy decision engine
  • ✓Audit event generation
  • ✓Local LLM inference serving
  • ✓Agent IDE and operations console

Why open? Enforcement must be auditable. If the software decides "allow" or "deny," you can read the code.

Commercial (Enterprise Edition)

Enterprise governance, administration, and compliance features for organizations at scale.

  • ✓Administrative control planes
  • ✓Policy lifecycle management
  • ✓Directory integrations (AD, Okta, LDAP, SAML)
  • ✓Compliance exports (SOC2, HIPAA, FedRAMP)
  • ✓Multi-tenant organizational governance

Why commercial? These components don't decide "allow vs deny" — they standardize operation at scale.

Lattice: Technical Architecture Overview
QUICK START

Up and Running in Three Commands

Install Lattice, start your headquarters, deploy your first agent.

# Install via Homebrew (macOS / Linux)
$ brew install latticehq/lattice/lattice
# Start Agent Headquarters
$ lattice server
# Deploy an agent with full governance
$ lattice create my-agent --template docker
# Serve local models (optional)
$ lattice inference serve --model mlx-community/Llama-3.2-3B-Instruct-4bit
# Open the Workbench (optional)
$ lattice workbench
Cryptographic Identity
Not a shared API key — a unique, verifiable principal.
Permission Controls
What your agent can and cannot do, enforced at runtime.
Immutable Audit Trail
Every action logged, tamper-evident, compliance-ready.
Local Inference
No API costs. Data never leaves your machine.
Web Dashboard
Monitor everything in the Workbench at localhost:3000.
Community Templates
Docker, K8s, AWS, GCP, Azure — deploy anywhere.
Browse ModulesExplore TemplatesDiscover Presets
WHO IT'S FOR

Same Platform. Every Scale.

From a weekend hack to Fortune 500 production — Lattice grows with you.

Weekend Hackers

brew install and go. Free local inference, no API bill, no credit card. Your agents run on your machine with full governance.

$ brew install latticehq/lattice/lattice

Startup Founders

Enterprise-grade governance from day one. Ship to enterprises without hiring a platform team. SOC2 and HIPAA compliance built in.

$ lattice create my-saas --template k8s

Enterprise Platform Teams

Unified governance across every cloud, every system. Self-hosted, auditable, vendor-neutral. Integrates with AD, Okta, LDAP.

enterprise@latticeruntime.com
GET IN TOUCH

Join the Community

Lattice is in active development across five open-source repositories. We welcome contributions, feedback, and early adopters.

GitHub

View the ecosystem, report issues, contribute code

Email

hello@latticeruntime.com

Security

Report vulnerabilities

Interested in Enterprise Edition?

We're working with select early partners to shape the enterprise features. Get in touch to learn more.

CONTACT ENTERPRISE TEAM