Easy install
Pick your agent stack.
AINL ships first-class paths for OpenClaw, ZeroClaw, and Hermes Agent. All three give you the compiler, Markdown importer, ecosystem shortcuts, and the ainl-mcp stdio server—wired the way each runtime expects (JSON MCP table, ZeroClaw MCP, or YAML mcp_servers on Hermes).
Three stacks—in plain terms
They are different assistant platforms, built by different teams. AINL is not a replacement for any of them: you install the platform you want first, then add AINL so your agent can run structured programs and workflows. The right choice is whichever platform you are actually using—or plan to install.
Choose OpenClaw if…
You follow the official OpenClaw setup from openclaw.ai, or you already run the openclaw command in your terminal. That path is a good default when docs or friends say “OpenClaw” without naming ZeroClaw.
OpenClaw is the mainstream, docs-first stack: the npm OpenClaw CLI, a workspace + skills layout under ~/.openclaw/, and the path most tutorials assume. AINL wires in through openclaw.json MCP entries and ainl-run. You can add optional bridge / cron automation when you want scheduled runs alongside chat. Choose it when you want the broadest official guidance and ecosystem overlap—not when you are optimizing for the smallest idle footprint.
Choose ZeroClaw if…
You use ZeroClaw from ZeroClaw Labs (or their open-source repo), or you already run the zeroclaw command. Same idea—different home for your agent—so match the guide to the tool you have.
ZeroClaw is written in Rust and is aimed at a smaller runtime footprint: less background weight on RAM and CPU, which makes it a practical choice on older laptops, compact home servers, and other setups where you want the assistant stack to stay lean while still speaking MCP and running structured workflows with AINL.
Choose Hermes Agent if…
You use Hermes from Nous Research as your agent runtime, or you already have ~/.hermes/config.yaml on your machine. Hermes has its own skills + learning-loop model and YAML-based MCP config—AINL wires in through ainl install-mcp --host hermes.
Hermes is a strong pick when you want a self-improving agent loop alongside deterministic AINL graphs. Once wired, you can emit hermes-skill bundles directly from your .ainl files and run them from the Hermes skill runner—no hand-written skill YAML required.
Still deciding? Open either walkthrough below; part one is only about getting that platform running. You do not need to understand the wiring details on the cards until you are ready.
OpenClaw
Official OpenClaw path—workspace-native skills, npm-based openclaw flows, and the integration story most community material references. New in v1.3.0: ainl install openclaw --workspace PATH sets up env, MCP, and crons in one command (use --dry-run to preview). Then ainl status (try --summary for one-line alerts) for a unified health view and ainl doctor --ainl for diagnostics. Full 5-minute flow: OpenClaw quickstart.
ZeroClaw
Rust-native host, MCP-first, and tuned for a lighter install—strong fit when you care about running on less powerful hardware without giving up tool access. Install the ZeroClaw skill or run ainl install-mcp --host zeroclaw to merge ~/.zeroclaw/mcp.json, add ainl-run, and expose MCP tools to your agent.
Hermes Agent
Official Hermes path—skills + learning loop from Nous Research. Run ainl install-mcp --host hermes to merge mcp_servers.ainl into ~/.hermes/config.yaml, add ~/.hermes/bin/ainl-run, then emit hermes-skill bundles for deterministic ainl_run execution.
CLI polish (OpenClaw, v1.3.0+)
Implemented in cli/main.py, documented in the quickstart, and mirrored in the repo README (subsection CLI polish). Bots can read tooling/bot_bootstrap.json → openclaw_commands.
| Command | What it does |
|---|---|
| ainl install openclaw | Merges env.shellEnv into workspace openclaw.json, bootstraps SQLite, registers gold-standard crons, restarts gateway — --dry-run previews. |
| ainl status | Unified health, budget (memory_records fallback), crons, drift, 7d tokens, est. cost avoided. Output: --json, --json-summary, --summary. |
| ainl doctor --ainl | Validates OpenClaw + AINL integration (env, schema, cron names, bootstrap flag). |
| ainl cron add | Wraps openclaw cron add with ainl run <path> — --cron or --every; --dry-run prints argv. |
| ainl dashboard | Serves tests/emits/server via scripts/serve_dashboard.py. Requires python3 scripts/run_tests_and_emit.py once from a git clone (creates server.py); the CLI errors with that hint if missing. --port, --no-browser. |
Project lock: aiNativeLang.example.yml → aiNativeLang.yml when missing. Shell shortcut: scripts/setup_ainl_integration.sh → ainl install openclaw.
MCP host integrations — one CLI pattern for every stack (ainl install-mcp --host …). Deep reference: OpenClaw integration (docs) · GitHub source · ZeroClaw integration (docs) · ZeroClaw (GitHub) · Hermes integration (docs) · Hermes (GitHub) · Hermes hub one-pager