The declarative language for AI agents

Define agents in YAML. Compile to signed, encrypted containers. Run on any compatible runtime. Version your tools. Protect your prompts.

name: research-assistant
version: 1.0.0

models:
  claude-sonnet-4:
    provider: anthropic
    options:
      temperature: 0.7

tools:
  - name: web_search
    version: ^1.2.0
  - name: web_fetch
  - name: artifacts

delegate:
  agents:
    - ./specialists/academic-researcher.yaml
    - ./specialists/market-analyst.yaml
  max_depth: 3
  permissions:
    academic-researcher: always
    market-analyst: session

What agentlang build does

$ agentlang build ./research-assistant

Resolving dependencies...
  tools: web_search@1.2.0, web_fetch@1.0.0, artifacts@2.0.1
  agents: academic-researcher@1.0.0, market-analyst@1.0.0

Generating lockfile...
  pinned 3 tools, 2 agents, 1 model
  hash: sha256:a1b2c3d4...

Packing container...
  bundled 6 files (14.2 KB)

Signing...
  key: agentlang-v1 (ed25519)

Encrypting prompts...
  algorithm: AES-256-GCM

Output: research-assistant@1.0.0.agent

The lockfile

version: "1"
resolved_at: "2025-01-29T10:30:00Z"

tools:
  web_search:
    version: 1.2.0
    hash: sha256:e5f6a7b8...
  web_fetch:
    version: 1.0.0
    hash: sha256:c9d0e1f2...
  artifacts:
    version: 2.0.1
    hash: sha256:a3b4c5d6...

agents:
  academic-researcher:
    version: 1.0.0
    hash: sha256:f7e8d9c0...
  market-analyst:
    version: 1.0.0
    hash: sha256:b1a2c3d4...

models:
  claude-sonnet-4:
    context_window: 200000
    supports_tools: true

files:
  agent.yaml: sha256:1a2b3c4d...
  prompt.yaml: sha256:5e6f7a8b...

Compared to other approaches

Install

# macOS / Linux
curl -fsSL https://agentlang.dev/install.sh | sh

# or with Homebrew
brew install agentlang/tap/agentlang

# verify
agentlang --version