Quickstart
Get started with AINL in under 3 minutes.
Pick the path that fits your setup. Most users start with Option A (plain pip) to explore the CLI, then move to Option B or C to wire it into an agent stack.
Option A · Recommended for 90% of users
Plain pip install
No agent stack required. Install the CLI, write a program, compile it, run it.
Verify it works
ainl --version
ainl validate --helpWrite your first program
Create hello.ainl using compact syntax:
# Your first AINL program — compact syntax
adder:
result = core.ADD 2 3
out resultAINL also supports a low-level opcode syntax for advanced use. Compact is recommended — 66% fewer tokens, same compiled output.
Validate and run
ainl validate hello.ainl
ainl run hello.ainlUse --strict for production-mode validation, --json-diagnostics for CI output.
Try a real example
# Clone the repo for demo programs
git clone https://github.com/sbhooley/ainativelang
cd ainativelang
ainl validate demo/monitor_system.lang
ainl run demo/session_budget_enforcer.langOption B · OpenClaw users
OpenClaw skill install
One-command AINL setup for OpenClaw — wires env, MCP, ainl-run, and gold-standard crons automatically.
Install AINL and run one-command setup (v1.3.0+)
pip install 'ainativelang[mcp]'
# Preview first:
ainl install openclaw --workspace ~/.openclaw/workspace --dry-run
# Then apply:
ainl install openclaw --workspace ~/.openclaw/workspaceMerges mcpServers.ainl and env.shellEnv into openclaw.json, installs ainl-run, and registers gold-standard crons idempotently.
Check health
ainl status
# Table: workspace, budget, crons, drift, 7d tokens, est. cost avoided
ainl status --summary
# One line for Slack / Telegram / CI
ainl doctor --ainl
# Integration diagnostics for OpenClaw + AINL
# Optional: schedule your own .ainl or open emitted dashboard
# ainl cron add ./workflow.ainl --cron "0 9 * * *"
# ainl dashboardGo deeper (optional)
For cap tuning, env profiles, and the full operator checklist:
OpenClaw gold standard guide →Option C · ZeroClaw users
ZeroClaw skill install
One-command bootstrap wired for ZeroClaw MCP and skill conventions.
Copy the skill
cp -r skills/zeroclaw/ ~/.zeroclaw/skills/ainl/Run the installer
cd ~/.zeroclaw/skills/ainl
./install.sh
# Or:
ainl install-mcp --host zeroclawWrites mcpServers.ainl into ~/.zeroclaw/mcp.json.
Verify
ainl --version
cat ~/.zeroclaw/mcp.json | grep ainlOption D · Hermes Agent users
Hermes Agent + AINL
Wire ainl-mcp into Hermes config, then emit agentskills-style bundles for deterministic ainl_run.
Install Hermes and AINL with MCP extras
Follow Hermes Agent until hermes works and ~/.hermes/ exists. Then:
pip install 'ainativelang[mcp]'
ainl install-mcp --host hermesEmit a skill bundle
ainl compile main.ainl --strict --emit hermes-skill -o ~/.hermes/skills/ainl-imports/my-worker/Hermes loads skills from ~/.hermes/skills/ and calls ainl_run for deterministic runs.
Verify
ainl doctor --host hermes
grep -A2 mcp_servers ~/.hermes/config.yamlFor Platform Teams
Start with recurring monitors and operational checks where cost compounds quickly. AINL compiles orchestration once, then runs deterministically without recurring LLM orchestration calls for each execution.
Strong fit for routine monitoring, digests, and policy-bound automations where predictable runtime behavior and lower recurring cost matter most.
For Enterprise Audit
Use strict validation before deployment, then enforce policy gates and capture JSONL execution tape during runtime for post-incident review and compliance trails.
This creates a clearer separation between workflow logic, runtime controls, and audit evidence than prompt-loop-only orchestration.
Common issues
- PEP 668On macOS or managed Linux hosts, pip may refuse to install globally. Use a venv:
python3 -m venv .venv && source .venv/bin/activate && pip install ainativelang - stale cacheIf you see an old
runtime_version, reinstall:pip install -U -e .or recreate the venv. - Python 3.9AINL requires Python 3.10+. Check with
python3 --version.