Unreal Labs
Unreal Labs explores how to run cloud agents efficiently, focusing on the harness, runtime, and inference surrounding models. It builds agent infrastructure and benchmarks/evals intended to make agents cheaper, faster, and more reliable.
At a Glance
- AI developers and engineering teams
- Companies deploying production cloud agents
- Research and science teams
- Software engineering and coding organizations
- +1 more
AI Tools by Unreal Labs
(1)
Unreal Agent
Open Source Go Agent Harness
Discussions
No discussions yet
Be the first to start a discussion about Unreal Labs
Latest News
Long-horizon agents at half the cost: context compaction, terminal UI, and Anthropic support for Unreal Agent
Unreal Agent v0.2.0 released
Unreal Labs introduced Unreal Agent, an open-source async-first harness claiming up to 40% cost savings versus Codex
Unreal Agent repository published under the MIT license
Products & Services
Open-source, async-first agent harness and SDK. It manages tool calls in the background, supports user steering while tools run, and includes a Go library, a runner executable, and a Harbor-compatible benchmark runner.
Custom benchmarks and evaluation environments for measuring agent quality, cost, latency, reliability, long-horizon performance, and agent trajectories on workflows, tools, and data supplied by customers.
Market Position
Unreal Labs positions its harness around asynchronous tool execution, low harness overhead, and cost efficiency rather than model novelty. Its published comparisons are primarily against Codex and Pi, while its technical discussion contrasts its own lightweight, provider-neutral approach with SDKs such as Claude Agent SDK; published benchmarks report similar or better pass rates at lower inference cost.
Leadership
Founders
Greg Tkachenko
Co-founder and CEO of Unreal Labs. Previously co-founded AI Factory, which was acquired by Snapchat for $166 million; his public profile also identifies him as a Sequoia scout/angel investor and co-founder of Tower.
Philipp Sinitsin
Co-founder and CTO of Unreal Labs; publicly described as working in machine learning and coding.
Oleg Yakubenkov
Co-founder of Unreal Labs and founder of GoPractice, Inc.; his public profile describes his role at Unreal Labs as product and analytics-oriented.
Executive Team
Greg Tkachenko
Co-founder and CEO
Previously co-founded AI Factory, acquired by Snapchat for $166 million; also identified publicly as a Sequoia scout/angel investor and Tower co-founder.
Philipp Sinitsin
Co-founder and CTO
Publicly described as working in machine learning and coding.
Founding Story
The team says it started from deploying agents in real products and observing that agents spend substantial time and tokens managing tool calls. Unreal Labs wanted a harness that could keep agents responsive and cost-effective, allow users to steer them without waiting, and run heterogeneous long-running tool work asynchronously.
Business Model
Revenue Model
The public site offers custom benchmarks, evals, long-horizon task work, evaluation environments, and agent-trajectory analysis through contact-led engagements. Unreal Agent itself is open source under the MIT license; no public self-serve pricing was found.
Target Markets
- AI developers and engineering teams
- Companies deploying production cloud agents
- Research and science teams
- Software engineering and coding organizations
- Organizations needing custom agent evaluation and benchmarking
- Production AI agents with long-running or heterogeneous tool calls
- Coding and software-engineering agents
- Scientific and research workflows
- Long-horizon, multi-step tasks requiring planning, tool use, recovery, and context compaction
- Benchmarking and evaluating agent configurations on realistic environments
- Teams building agent-first products that need provider compatibility and lower inference cost