Compile once · Run forever · Open core
Turn vague LLM conversations into deterministic, auditable production workers. AINL fills the runtime-shaped hole in the AI stack: great models, great tools — no reliable execution layer until now. Early adopters report 2–5× lower recurring token spend on high-frequency workflows.
See the 3-minute quickstart guide → — pip install, scaffold, run, visualise.
View real benchmarks — LangGraph & Temporal sizing from committed CI JSON.
Read our growth plan — milestones, risk mitigations, and community-first scaling.
Validation transparency deep dive — reachability checks, strict diagnostics, and JSONL execution tape.
Join the community: Discussions · Telegram · @ainativelang
# Compact syntax — Python-like, 66% fewer tokens
email_monitor:
in: inbox_name
last_check = cache.GET state "last_email_check"
emails = email.G inbox_name
email_count = core.len emails
if email_count > 5:
out "notify"
out "ok"Community token (pump.fun)
Read the technical whitepaper (v1.4.1) — graph IR, adapters, benchmarks, and operator model.
Try in 3 minutes
pip install ainativelang ainl init my-first-worker cd my-first-worker ainl validate main.ainl --strict ainl run main.ainl ainl visualize main.ainl --output graph.mmd ainl serve --port 8080 # optional: HTTP API
The generated main.ainl includes inline comments explaining L1: (label), R cache get (request from cache adapter), J (join/return), and branching — designed for newcomers, production-ready from day one.
Why AINL
Teams using AINL turn workflows into compiled graphs — not fragile prompt loops — so agents behave like real infrastructure. Early adopters report clearer cost predictability and faster iteration on recurring, monitoring-style workloads.
Orchestration lives in a compiled graph IR, not in the model. The same workflow produces the same result every time — inspectable, diffable, auditable.
Author with an LLM once, compile, then run without re-invoking the model. Cuts recurring token spend by 2–5× on high-frequency workflows.
First-class MCP server for AI IDEs. Skill installs for OpenClaw and ZeroClaw wire the compiler, runner, and bridge in one command.
pip install ainativelang then ainl init my-worker. Scaffold, validate, run, visualise — no clone required.
How AINL compares
Card 1
| Aspect | Prompt loops | AINL |
|---|---|---|
| Recurring orchestration tokens | High — every run | 0 after compile |
| Authoring cost | Large prompts each time | ~99 tokens (hybrid) |
| Monthly cost (10 workflows) | $100–$2,000+ | Near $0 |
Card 2
| Feature | Typical AI tools | AINL |
|---|---|---|
| Determinism | Low–Medium | Guaranteed (compiled IR) |
| Validation | Manual / none | Strict compiler + JSON diagnostics |
| Auditability | Limited | Policy gates + JSONL execution tape |
Card 3
| Experience | LangGraph / Temporal | AINL |
|---|---|---|
| First workflow | Hours–days | < 3 min with ainl init |
| AI IDE integration | Partial | Native MCP + OpenClaw |
| Single source of truth | No | Yes — emit multiple targets |
AINL numbers from committed CI benchmarks (March 2026). View full comparison tables → · LangGraph → AINL migration guide →
For Platform Teams
AINL is a strong fit for cost-sensitive routine monitoring and operational digests: compile orchestration once, validate it, then run deterministic workflows repeatedly with explicit adapter boundaries.
For Enterprise Audit
Use strict compiler checks before deploy, then enforce capability/policy gates and capture JSONL execution tape for compliance, incident review, and operator trust.
See the production tape replay example and how to bundle JSONL evidence for auditors in our SOC 2 alignment checklist.
We offer commercial support and a hosted runner for teams that need guaranteed SLA and dedicated help — see COMMERCIAL.md for support tiers and how they address operational backing alongside open core.
Enterprise assets
Who AINL is for
AINL is built by AI engineers, for teams who need orchestration to behave like infrastructure — not a black-box prompt loop — with strict validation, policy gates, and bridges into OpenClaw, Hermes Agent, ZeroClaw, Nemoclaw, and other MCP-native hosts.
AINL is less optimized for casual chatbot prototyping, one-off prompt hacks, or no-code-first usage.
AINL matters when AI moves beyond demos and starts doing real operational work. It helps reduce the gap between “a clever AI demo” and “a workflow a business can actually run” — repeatable, inspectable, stateful, tool-using, cost-aware, and easier to validate and maintain.
An AINL program gives an agent explicit control flow, explicit side effects, compile-time validation, capability-aware boundaries, externalized memory, and reproducible runtime behavior. The LLM stops being the whole control plane and becomes a reasoning component inside a deterministic system.
Community in Action
Real teams are already using AINL to:
Recent highlights:
We're actively working toward our 90-day goals from the Growth Plan: more external contributors, public case studies, and the first community maintainer.
Want to be featured?
Share your workflow in GitHub Discussions or on Telegram. The best examples will be highlighted here and on our monthly community spotlights.
Built with ❤️ by the growing AINL community — not just one person.
Built with AINL
Real programs, measurable outcomes, and links to source or reports. More entries monthly — see the spotlights log on GitHub.
We actively feature user-submitted workflows. Submit yours in Discussions #14.
Discussions are now live — jump in: #14 Share your first workflow, #15 LangGraph → AINL, #16 Enterprise audit. Hub: welcome (#13). Canonical copy: DISCUSSIONS_POST_EXACT.md.
Showcase
External builder
--trace-jsonl.Independent
vector_memory.UPSERT + SEARCH, gated by an explicit ops budget branch — strict validation, no runtime orchestration LLM.Early adopter
crm_db; no runtime orchestration LLM.Compliance team
audit:policy_violation or audit:within_policy, and full JSONL execution tape for SOC 2 evidence bundles.Share yours → Submit a workflow for the next spotlight.