Lago

Event-sourced persistence layer for long-lived AI agents.

Lago consolidates all agent state changes — tool use, file writes, messages, memory — into a single event-sourced, versioned system with streaming I/O.

Features

• Event sourcing — All state derived from an append-only journal
• Content-addressed blobs — SHA-256 hashing with zstd compression
• Filesystem branching — Git-like branching and diffing for agent workspaces
• gRPC streaming ingest — Bidirectional streaming via tonic
• HTTP REST + SSE — Axum-based API with Server-Sent Events
• Multi-format SSE — OpenAI, Anthropic, Vercel AI SDK compatible
• Policy engine — Rule-based tool governance with RBAC
• Embedded storage — redb (ACID, pure Rust, zero external dependencies)

Architecture

lago-cli / lagod           Binaries (CLI + daemon)
  ├── lago-api             HTTP REST + SSE streaming (axum)
  ├── lago-ingest          gRPC bidirectional streaming (tonic)
  ├── lago-policy          Policy engine + RBAC
  ├── lago-fs              Filesystem manifest + branching
  ├── lago-journal         Event journal (redb-backed)
  ├── lago-store           Content-addressed blob storage
  └── lago-core            Foundation types, traits, errors

Installation

From GitHub Releases

curl -fsSL https://raw.githubusercontent.com/broomva/lago/main/install.sh | bash

From Source

cargo install lago

From crates.io (library crates)

[dependencies]
lago-core = "0.1"
lago-journal = "0.1"
lago-store = "0.1"

Quick Start

# Start the daemon
lagod --config lago.toml

# Create a session
lago session create --name my-agent

# List sessions
lago session list

# Stream events via SSE
curl -N http://localhost:3000/v1/sessions/<id>/events

Development

# Build
cargo build --workspace

# Test (295 tests)
cargo test --workspace

# Lint
cargo clippy --workspace

# Format
cargo fmt --all

# Check dependencies
cargo deny check

Crates

License

MIT