PRODUCT

The runtime your AI agents deserve

Alius is purpose-built for running AI agents in production. Every feature is designed to make agents more reliable, observable, and controllable.

Install Anywhere

Get up and running in seconds. Choose your preferred installation method.

curl (macOS / Linux)

curl -fsSL https://alius.tech/install | sh

npm

npm i -g @alius-tech/alius

Homebrew

brew install alius
EXECUTION MODEL

Session · Run · Trace

The three-level execution model is the backbone of Alius. Sessions provide persistent context, Runs execute structured plans, and Traces record every detail.

  • Stateful session management with context persistence
  • Structured plan execution with step-level tracking
  • Complete audit trail for every agent action
// alius.config.toml
[session]
persist = true
max_runs = 100
[run]
timeout = "5m"
retry_on_failure = true
[trace]
level = "full"
export = ["json", "otel"]
LLM Provider
OpenAI, Anthropic, Local
Tool Interface
Shell, File, HTTP, Custom
Memory Interface
Context, Episodic, Semantic
PROTOCOL INTERFACE

Pluggable by Design

Protocol interfaces decouple your agent logic from specific providers. Swap LLM providers, tool backends, or memory systems without changing your agent code.

  • Provider-agnostic LLM integration
  • Extensible tool and capability system
  • Custom protocol implementation support
SECURITY

Shell Gate

Not every action should run autonomously. Shell Gate is the human-in-the-loop security layer that requires explicit approval for sensitive operations like shell commands, file writes, and network requests.

  • Granular permission controls per operation type
  • Auto-approve rules for trusted operations
  • Full audit log of all approvals and denials
shell gate
⚡ Shell Gate: Approval Required
Command: rm -rf /tmp/build-cache
Risk: MEDIUM
Agent: deploy-agent
✓ Approve ✗ Deny ⏳ Auto (5m)
3
Contexts
12
Episodes
47
Facts
ctx project-layout 2h ago
epi debug-auth-flow 1d ago
fac uses-postgres-15 3d ago
MEMORY SYSTEM

Agents That Remember

Alius agents persist knowledge across sessions through a three-tier memory system: Context (working memory), Episodic (experience memory), and Semantic (factual memory).

  • Context: short-term working memory per session
  • Episodic: learned experiences from past runs
  • Semantic: extracted facts and knowledge base