YLEM LABS, INC. · SAN FRANCISCO · ASTANA

The full stack for autonomous AI

Ylem Labs is an AI company. We research world models, build the harness that runs agents, and ship the infrastructure that connects, constructs and protects them — one ecosystem, designed together.

Explore the ecosystem · Read our research

Five products. One operating system for agents.

Every layer an agent needs — from the model it thinks with to the firewall it passes through — built by one team, on one set of primitives, with one API.

Harness · agent runtime · R&D preview

The execution layer for long-running agents: plan/act loops, typed tools, memory, skills, sandboxes and background jobs on a Pydantic-typed core.

plan / act / verify loops
skills, MCP, sandboxes
background and scheduled jobs

World · JEPA-style world model · research

Self-supervised models that predict in latent space instead of tokens — so agents can plan over consequences, not just text.

joint-embedding predictive architecture
latent planning for agents
open research releases

Gateway · model gateway · live

One OpenAI-compatible API to 300+ models with live routing, fallback, caching and batch.

failover under 300 ms
semantic cache
per-team budgets

Box · agent builder · beta

Compose agents from prompts, tools, knowledge and MCP servers, then deploy them in your cloud in one click.

visual and code builder
RAG and knowledge bases
on-prem deployment

Guard · AI firewall · pilots

Policy-as-code security for every model call: PII, secrets, document leakage and prompt injection — with our own guard model.

monitor, then calibrate, then enforce
in-house guard model
DLP for documents

Designed together, not bolted together.

Agents are built in Box, run in Harness, reach models through Gateway and pass Guard on every call. Swap any layer for your own — the interfaces are open.

One request: your apps, then Box, then Harness, then Gateway, then World or 300+ models from 40+ providers. The return path is traced, budgeted, and guarded.

Models that predict the world, not the next token.

Our research line explores JEPA-style world models: encoders that learn to predict missing parts of the world in latent space. It is the foundation we want our agents to plan with.

October 2026 — Latent planning for tool-using agents
September 2026 — Harness > Model: how much of agent quality comes from scaffolding
August 2026 — Tunneling Fallback: latency-bounded routing across providers
July 2026 — Cascaded guard models at the gateway

The lab in numbers

5 products, one stack
300+ models on Gateway
100+ engineers and researchers
2 R&D hubs: San Francisco and Astana

How we build

Own the stack. Every layer that touches an agent — model, runtime, gateway, firewall — is ours to measure and improve.

Ship research. A paper is not done until it runs in production. Research and product share one roadmap.

Safety at the gateway. Autonomy is earned with policy, not prompts. Every call is inspected before it leaves your boundary.

Help us build the agent stack.

We're hiring researchers and engineers in San Francisco and Astana, and partnering with companies that want to run agents in production. Write to sales@ylemlabs.ai.

ylem /ˈaɪləm/ n. — the primordial matter from which all elements formed (Gamow & Alpher, 1948).