Submilli Runtime

Submilli Runtime is a code-execution runtime that runs agent-generated TypeScript in WebAssembly with semantic permissions.

Submilli is a code-execution runtime designed for business agents that generate code. The agent submits TypeScript code, which the runtime executes in WebAssembly for isolation. It features a semantic permission model defined in YAML blueprints, allowing owners to set precise guardrails like refund limits based on customer tiers. Packages wrap APIs and store credentials so generated code never sees secrets. It supports integrations like GitHub, Slack, Gmail, Google Drive, Linear, Notion, Sentry, and web search, plus MCP servers.

Submilli Runtime at a glance

Best for: Software engineer
Runs on: macOS, Linux, Windows, API, CLI, self-hosted, open source
Pricing: Free
Released: October 2026
Category: AI coding tool
Website: github.com/submilli/submilli-runtime

Key features

Executes agent-written TypeScript code in WebAssembly
Enforces semantic permission rules in YAML blueprints
Hides API credentials inside curated packages
Limits memory, execution time, and stack depth
Maintains an audit trail of every run and refusal
Supports Model Context Protocol servers as packages
Provides HTTPS, API tokens, and encrypted secret storage

Use it for

Execute agent-generated TypeScript securely in WebAssembly
Define semantic guardrails for agent actions in YAML
Isolate code execution away from node:http or node:fs
Audit every refusal, run, session, and admin change

Alternatives to Submilli Runtime

OpenSandbox — Provides isolated environments with multi-language SDKs and lifecycle management.
SmallCode — Runs fully local with budget-managed context and a forgiving parser for small models.
Warp — AI terminal plus cloud agents with built-in evals and benchmarks
Daytona — It provisions sandboxes in under 90 milliseconds with full security isolation.
Pi Subagents Lite — Spawn custom agents in isolated sessions with minimal token overhead.
jevgrep — Reduces coding agent costs by providing targeted, relevant repository context.
Wasi Python — Runs untrusted AI-generated code safely inside an isolated WebAssembly sandbox.
Lloyal — Builds apps with embedded models that run offline without user API keys.
Weft — It enables safe parallel coding agent work without file locks.
Memdoor — Runs repeatable, headless workflows with local keys and proof for every step.
Melra — Governs AI agent tool calls with local policies, approval gates, and verified receipts.
AgentGrid — Centralize multiple AI coding agents and development tools on one interactive canvas.

All Submilli Runtime alternatives