Neander is a minimalistic, statically-typed language designed exclusively for AI agents to generate and execute small programs that implement API-calling logic for enterprise systems.
The language name references the Neanderthal: a simple, predictable, discoverable design from an earlier era — adapted to the modern world.
Programs are fully self-contained, stateless artifacts that can be transmitted from client to server and executed safely without external sandboxing — the safety guarantees are embedded in the language itself. Instead of loading a sprawling catalog of tool definitions into an agent's context, Neander gives the agent a single compact language reference: the agent discovers the functions it needs and calls them by writing programs.
In short, Neander is:
- Interpreter-only — no compiler, no bytecode.
- Safe-by-construction — no sandbox required.
- Not Turing-complete by design — programs are guaranteed to terminate.
- Stateless — each program is a complete, independent artifact.
- For orchestrating externally-registered APIs: discovery, conditional logic, iteration, and error handling. The language defines no APIs of its own; the runtime hosts APIs registered by the embedding application.
This repository contains two documents, each kept as an editable golden source alongside the latest published revision:
| Document | Golden source | Latest publication |
|---|---|---|
| Neander Language Specification — the normative definition of the language | spec/neander.md |
spec/neander-v1_draft_20260622T151500Z.md |
| Neander Reference — a practical, example-driven guide for writing programs aimed at AI agents | spec/neander-reference.md |
spec/neander-reference-v1_draft_20260618T104500Z.md |
Read the latest published revision. Edit the golden source. Published revisions are immutable, timestamped snapshots and must not be modified.
This work is licensed under the CC BY 4.0. See LICENSE for the full text.
See CONTRIBUTING.md.