The agentic operating system

Built for agents, down to the hardware.

qOS is HQ's operating system, written in Rust. It runs state-of-the-art agents on hardware-isolated microVMs that cold-boots under 100ms, on dedicated hardware we run ourselves. Fully isolated and secure.

The runtime

Every agent gets intelligent.

Each thread runs in its own microVM: true hardware isolation with access to thousands of tools and skills. They are dispatched across low-latency datacenters on hardware HQ maintains end to end, so nothing noisy or untrusted shares your agent's kernel.

~100msmicroVM cold boot
Hardwareisolation, not a shared kernel
Sub-msintra-datacenter latency
Dedicatedhardware we run ourselves

Polyglot by design

Runs three SDKs. Pick the model. Run as many as you want.

Runtimes

Claude Agent SDKAnthropic's official agent runtime
OpenAI Agents SDKOpenAI's official agent runtime
OpenCode SDKOpen runtime, bring any model

Models

Claude Opus 4.8Claude Sonnet 4.6OpenAI GPT-5.5Grok 4.3 via OpenCode

Give each agent a handle and a specialty. Run one, four, or a fleet, each on the SDK and model that fits its job. We track all the major model families as they ship.

Inside the model layer ›

Tools, skills, integrations

A real Linux machine, fully the agent's.

Agents reach a broad range of tools through MCP and locally installed CLIs and skills, all in one safe environment. Tailor the skills, toggle integrations on or off, and let an agent install anything else it needs. It just works, like a normal Linux server, 100% the agent's.

  • MCP-native. A common tool protocol across every agent and integration.
  • Shell + filesystem. Local CLIs are first-class, not wrapped per tool.
  • Composable skills. Enable, disable, or tailor per agent.
  • Self-extending. An agent can install the tools it needs, in its own sandbox.

Dedicated core systems

We built the ground qOS runs on.

Search, cache, storage, email, graph, browser, documents, extraction. Our own systems, used in production for large-scale ingestion, so agents never wait on someone else's infrastructure. Written in Rust for performance and robustness.

Torque Search

Typesense-compatible search, rebuilt in Rust.

  • In-memory core, mmap-backed persistence
  • Hybrid BM25 + vector (HNSW), RRF fusion
  • 6 to 19x faster than Typesense on real data (CPU)
Explore Torque ›
Forge Cache

A multi-threaded, microsecond-class in-memory store.

  • Redis-class structures, no single-threaded ceiling
  • Binary TCP, UDP and shared-memory transports
  • WAL + snapshot persistence
Explore Forge ›
Trove Storage

S3-compatible object storage in one masterless binary.

  • AWS SigV4 S3 API, works with boto3 and rclone
  • Small-file optimized: O(1) reads
  • Replication, erasure coding, self-healing
Explore Trove ›
Sentio Email

A modern, API-first email platform (MTA).

  • Inbound and outbound SMTP, MX + TLS
  • DKIM, SPF, DMARC, ARC, DANE, MTA-STS, BIMI
  • Three-tier anti-spam with an LLM tiebreaker
Explore Sentio ›
Traverse Graph

An in-memory graph database, Rust-native.

  • openCypher and ISO GQL over one executor
  • Neo4j-compatible Bolt, gRPC, built-in MCP server
  • Hot-swappable databases, lock-free readers
Explore Traverse ›
Mirage Browser

A custom-built web browser, driven from Rust.

  • A real browser we built, not an automation shell
  • Human-like input, per-session proxy rotation
  • Locale, timezone and geo pinned to the egress
Explore Mirage ›
Lector Documents

A native-quality PDF engine, in the browser.

  • Direct-to-canvas rendering in a Web Worker
  • Annotations, forms, redaction, signature reading
  • Text extraction and regex search
Explore Lector ›
Sift Extraction

ExifTool and Poppler, replaced by one Rust library.

  • Image metadata: EXIF, IPTC, XMP, ICC (JPEG to RAW)
  • PDF text, image and metadata extraction
  • Embeddable in five languages
Explore Sift ›

Put a fleet of agents on qOS.

See HQ running in your own Slack or Teams, on the operating system we built for agents.

Request access