HYRO
Agent OS · v0.1.0

HYRO Cloud
The Operating System for Autonomous Agents

A terminal-first cloud runtime to create, deploy, execute and monitor autonomous AI agents — with tools, persistent memory and native MCP connectivity.

01/Observe02/Decide03/Execute04/Remember
Install CLI@HyroCloud
$npm install -g hyro
hyro — 0.1.0 — zsh

HYRO TERMINAL·v0.1.0

hyro ❯

// features

Everything an agent needs to ship

From the terminal to durable cloud runs — HYRO is the full stack for autonomous software.

01

Terminal-first CLI

A hacker-grade blue terminal with a live REPL and one-shot commands. The CLI is the product.

02

Cloud runtime

Agents run as durable, observable runs. Every step is streamed over SSE and fully replayable.

03

Native MCP

Install Model Context Protocol servers; HYRO discovers tools and exposes them — deny-by-default.

04

Memory engine

Facts, goals, preferences, conversations and state in PostgreSQL + pgvector with semantic recall.

05

Multi-model

Anthropic, OpenAI, Gemini and OpenRouter behind one interface. Switch with a single command.

06

Marketplace

Publish, discover and install agents — Research, Crypto, Growth, Trading and Builder out of the box.

// mcp

Native Model Context Protocol

Install servers, discover tools, grant permissions, and execute — all from the same HYRO terminal you use to run agents.

# Grant github tools to research-agent

hyro ❯ mcp install @hyro/github

✔ Installed github-mcp v1.2.0

hyro ❯ mcp grant research-agent github__search_issues

✔ Granted 1 tool · deny-by-default for others

hyro ❯ mcp list --agent research-agent

→ github__search_issues

memory_search · think · (builtin)

01

Install

Pull signed MCP packages from the HYRO registry or point at a local server.

02

Discover

Handshake exposes tools/resources; HYRO catalogs them per server with version pins.

03

Grant

Deny-by-default. Grant tool-level permissions per agent before any call is proxied.

04

Execute

Runtime proxies tools/call with timeouts, audit logs, and streamed step events.

// memory

Persistent agent memory

Facts, goals, preferences, conversations and state — stored in PostgreSQL with pgvector semantic recall, scoped per agent.

Memory types

::fact::goal::preference::conversation::state
  • →Cosine ANN retrieval re-ranked by importance and recency
  • →Export / import portable JSONL for backups and migrations
  • →Works offline with a local embedder — zero API keys required
memory_search · pgvector · agent:research

hyro ❯ memory search "Q3 launch blockers"

[fact] score 0.91 · API rate limits blocking staging deploy

[goal] score 0.87 · Ship public beta before Oct 1

[state] score 0.84 · Last triage: 3 issues labeled P1

3 results · 12ms · hybrid rerank

// cli

The CLI is the product

A hacker-grade terminal with a live REPL and one-shot commands. Run offline with zero keys, or connect to HYRO Cloud for durable execution.

hyro loginAuthenticate with HYRO Cloud
hyro run "<task>"Execute an agent run (offline or cloud)
hyro agents listList agents in your workspace
hyro model use <id>Switch the active LLM provider/model
hyro mcp install <pkg>Install an MCP server from the registry
hyro memory search <q>Semantic recall across agent memory
Read CLI docs

# Quick start — works offline

$ npm install -g hyro

$ hyro

→ Interactive REPL when attached to a TTY

$ hyro run "summarize today's standup notes"

✔ offline · 4 steps · local memory

Observe · Decide · Execute · Remember

// architecture

Four layers, one runtime

Terminal-first interface, durable orchestration, MCP connectivity, and vector memory — composed as a single agent operating system.

Layer 1

Interface

  • Blue CLI & REPL
  • TypeScript SDK
  • REST + SSE API

Layer 2

Runtime

  • Run orchestrator
  • Tool sandbox
  • Step streaming
  • Offline local mode

Layer 3

Connect

  • MCP Hub
  • Provider router
  • Webhook ingress
  • Agent marketplace

Layer 4

Persist

  • PostgreSQL
  • pgvector recall
  • Redis queue
  • Run replay store
┌─────────────┐     SSE      ┌──────────────┐     MCP      ┌─────────────┐
│  HYRO CLI   │ ───────────► │   Runtime    │ ◄──────────► │  MCP Hub    │
│  + SDK      │   REST API   │  Orchestrator│   tools/call │  Registry   │
└──────┬──────┘              └──────┬───────┘              └─────────────┘
       │                            │
       │                     ┌──────▼───────┐
       │                     │  PostgreSQL  │
       └────────────────────►│  + pgvector  │
                             │  + Redis     │
                             └──────────────┘

// roadmap

Shipping in public

Foundation is live. Runtime hardening, MCP GA, and marketplace teams are next on the roadmap.

P0Shipped

Foundation

Monorepo, typed kernel, Fastify API, pgvector schema, HYRO CLI, landing page.

P1Planned

Runtime hardening

Durable run queue, full SSE streaming, live provider completions, per-run caps.

P2Planned

MCP Hub GA

Hosted signed registry, sandbox profiles, interactive permission grants.

P3Planned

Memory engine

Hybrid retrieval, namespaces, summarization & compaction jobs.

P4Planned

Deploy & schedule

Callable agent endpoints, webhooks, cron & event triggers, versioned rollouts.

P5Planned

Marketplace & teams

Publishing pipeline, revenue share, orgs, SSO, audit logs, enterprise scale.

// faq

Questions from the terminal

Everything you need to know before your first `hyro run`.