SQUID Language Stack

A multi-language AI execution system.

Languages

• SQUIDDY → intent layer
• INK → schema layer
• TENTA → workflow layer
• KRAK → runtime optimization
• TIDE → event system

Goal

Turn natural language into deterministic, executable AI workflows.

Pipeline

SQUIDDY → INK → TENTA → KRAK → TIDE → Execution

CLI

Build and run:

npm run build
npm start

Run any supported file:

node dist/cli/index.js <file> [schema.ink]

| Extension | Description | |-----------|-------------| | .squiddy | Full pipeline: lex → parse → validate → INK → TENTA → KRAK → execute → SQUIDDY runtime | | .tenta | Explicit flow: parse → KRAK → INK → execute | | .tide | Event schedule: parse → trigger registered flows → execute |

Optional second argument: INK schema path (default: examples/schema-example.ink).

Examples

node dist/cli/index.js examples/hello-world.squiddy
node dist/cli/index.js examples/multi-agent.squiddy
node dist/cli/index.js examples/simple-flow.tenta
node dist/cli/index.js examples/failure-route.tenta   # exits 1 (failure demo)
node dist/cli/index.js examples/event-example.tide

Verify

Run the full regression suite:

npm run verify

Build phases

See docs/build-plan.md. Phases 1–5 are implemented.

Phase 5 — LLM integration (simulated)

The LLM layer is compile-time only: it maps known natural-language intents to valid SQUIDDY agents, validates them through the existing pipeline, and can generate training datasets from execution traces.

# Generate a SQUIDDY agent from an intent and run it
node dist/cli/index.js --llm "create agent that logs hello"

# Run a .squiddy file and export execution trace dataset
node dist/cli/index.js --dataset examples/multi-agent.squiddy

# Simulate fine-tuning on a JSONL dataset
node dist/cli/index.js --finetune examples/datasets/execution-trace.jsonl

AI integration (Cursor, Claude, Kimi, ChatGPT)

Make AI models know and use your languages:

| Layer | What | |-------|------| | 1 | AGENTS.md + .cursor/rules/ — context | | 2 | MCP server — squid_* tools (compile, run, verify) | | 3 | VS Code extension — syntax, snippets, Run | | 4 | Fine-tune guide — dataset + training prep |

Full setup: docs/ai-integration.md

npm run build        # required before MCP parse/run tools
npm run mcp          # MCP server (stdio, for Cursor)