Roadmap

Prioritizing agent-first capabilities. See the story for why.

Phase 1: Foundation

Capability Status
Bootstrap compiler (C), Lexer, Parser, Codegen✓ Done
LLVM IR generation, native compilation via clang✓ Done
User-defined functions, Math module✓ Done
Control flow (if/else, loops, counters)✓ Done
Implicit main, .env file support✓ Done

Phase 2: Agent Capabilities

Capability Status
HTTP module - GET, POST, PUT, DELETE, PATCH✓ Done
HTTP custom headers, Bearer auth, Basic auth✓ Done
LLM module - Claude API calls✓ Done
JSON parsing and manipulation (cJSON)✓ Done

Phase 3: MCP Client

Capability Status
MCP Tools - list and use tools✓ Done
MCP Resources - list and read resources✓ Done
MCP Prompts - list and get prompt templates✓ Done
MCP Logging - set log levels✓ Done
Protocol version 2025-03-26✓ Done

Phase 4: Enhanced Capabilities

Capability Status
OAuth 2.1 for MCP authenticationComing next
SSE / HTTP streaming for real-time responsesComing next
OpenAI API support (GPT-4, etc.)Coming next
String operations (concat, split, extract)Coming next

Future Exploration

Capability Status
Conversation state / multi-turn dialogsExploring
Human View generator (NERD → readable code)Exploring
Context storage, long-term memoryExploring
WASM target for browser/edgeExploring
IDE support (syntax highlighting, LSP)Exploring

Philosophy

We're not chasing feature parity with general-purpose languages. NERD is built for LLMs, not for human authorship. We're building what agents need first, then expanding based on gaps we discover.

Read more: LLM-Native Language Going Agent-First

Contribute

See CONTRIBUTING.md.