Your agents crash silently. You can't prove what happened. Lattice fixes both.
AI agents crash and nobody knows. When something goes wrong, there's no audit trail. The status quo is systemd plus prayer. Every tool in this space helps you write agents. None of them help you run agents safely.
Lattice Runtime is a single Go binary that gives your agents crash-proof execution (embedded Temporal), cryptographic audit (SHA-256 hash-chained), budget enforcement, and zero-trust mesh networking. Install it, deploy your agents on top. No code changes required.
brew install latticehq/lattice/lattice
lattice server| Component | What it does | License |
|---|---|---|
| Runtime | Crash-proof runtime — identity, auth, audit, budget, mesh | Apache 2.0 |
| Workbench | 316K-line multi-model agent workspace | MIT |
| SDK | Go SDK for building on Lattice | Apache 2.0 |
| Inference | Local AI serving — MLX on Apple Silicon | Apache 2.0 |
| Operator | Self-hosted deployment via SSH | Apache 2.0 |
| Registry | Community modules, templates, stacks | Apache 2.0 |
| Terraform Provider | Infrastructure as code | MPL 2.0 |
| Toolbox | macOS app manager | MIT |
| Homebrew | One-line install | MIT |
- Crash-proof — Embedded Temporal. Agents survive failures. State is never lost.
- Auditable — SHA-256 hash-chained audit trail. Tamper with one record, every subsequent hash breaks. SOC2/HIPAA/FedRAMP as a side effect.
- Governed — RBAC + ABAC via Rego. Budget enforcement. Tool whitelisting. PII blocking. All at runtime.
- Open source — Apache 2.0. If software decides "allow" or "deny," the decision logic must be auditable.
- Self-hosted — Your data stays on your infrastructure. Cloud, on-prem, edge, air-gapped.
- Website: latticeruntime.com
- Vision: VISION.md
- Security: security@latticeruntime.com
I killed the agent. It came back. I can prove it.