LithosAI

Cover Block

Attribute	Value
Company Name	LithosAI
Tagline	Open-source agent-serving platform that improves AI agents from production traces
Stage	Pre-Seed
Business Model	Open Source / Commercial
Industry	Other
Technology	AI / Machine Learning
Geography	North America
Founding Team	Academic Spinout (Dimitrios Skarlatos, Zhihao Jia)

Links

• Website: https://www.lithosai.com/
• GitHub: https://github.com/lithos-ai/motus

What an Investor Needs First

LithosAI is building an open-source platform for deploying and improving AI agents using signals extracted directly from production [LithosAI.com, 2026]. The company's core project, Motus, is a model-agnostic serving layer that allows developers to orchestrate across foundational models without vendor lock-in [SOTA Sync, 2026]. This venture is an academic spinout, founded by Carnegie Mellon University professors Dimitrios Skarlatos and Zhihao Jia, whose research in computer systems and machine learning forms the technical bedrock of the company [CMU SCS News, 2026].

The business model follows an open-source-first path, with Motus released under an Apache 2.0 license [LINUX DO, 2026]. The company is in a pre-seed, pre-revenue stage.

Data Accuracy: YELLOW -- Core product and team claims are sourced from the company's own materials and founder profiles; financial and traction data is unconfirmed.

Taxonomy Snapshot

Axis	Classification
Stage	Pre-Seed
Business Model	Open Source / Commercial
Industry	Other
Technology	AI / Machine Learning
Geography	North America
Founding Team	Academic Spinout

Inside the Company

LithosAI is an academic spinout from Carnegie Mellon University, founded by professors Dimitrios Skarlatos and Zhihao Jia [LithosAI.com, 2026]. The company's public presence centers on the release of its open-source project, Motus, in 2026 [GitHub, 2026].

The founding narrative is rooted in the founders' research at Carnegie Mellon. Zhihao Jia's doctoral work, which earned the Arthur Samuel Best Doctoral Thesis Award, focused on automated discovery for machine learning systems [CMU SCS, 2026]. This foundational research, alongside a Meta Research Award for work to reduce the costs of AI techniques, informs the technical thesis behind LithosAI's platform [CMU SCS News, 2026].

Data Accuracy: YELLOW -- Company website and founder academic pages provide corroboration; corporate details are absent from public databases.

Under the Hood

The core of LithosAI is Motus, an open-source platform designed to serve and improve AI agents using data from their own production runs. The product orchestrates across any underlying model while extracting improvement signals from operational traces like task outcomes, latency, and cost [LithosAI.com, 2026].

Motus follows a "no-framework principle," allowing developers to use it without modifying existing agent code written for popular SDKs [SOTA Sync, 2026]. The platform is released under the Apache 2.0 license [LINUX DO, 2026]. Its workflow allows a single command to serve an agent locally or deploy it to a managed cloud service [SOTA Sync, 2026]. Motus's multi-model orchestration achieved a 79% score on the SWE-bench coding benchmark while halving associated costs [LINUX DO, 2026].

Data Accuracy: ORANGE -- Claims are sourced from the company's own materials and technical blog posts; independent verification of performance benchmarks and deployment capabilities is not available.

Competition and Substitutes

LithosAI enters a crowded market for AI agent tooling, but its positioning as an open-source, model-agnostic platform for learning from production data carves a distinct wedge.

Incumbents are dominated by vendor-specific SDKs like OpenAI's Agents SDK, Anthropic's SDK, and Google's ADK. Motus claims compatibility with these, following a "no-framework principle" [SOTA Sync, 2026]. Direct challengers include open-source agent orchestration platforms like LangChain's LangGraph and AutoGPT. Adjacent substitutes include commercial agent deployment platforms like Google's Vertex AI Agent Builder or AWS Bedrock Agents, which offer managed serving but typically introduce vendor lock-in.

LithosAI's defensible edge is technical and talent-based. The founding team's academic research in systems and machine learning provides a foundation for the core technical challenge of efficient, multi-model orchestration [CMU SCS News, 2026]. The performance claim that Motus's multi-model orchestration achieves 79% on SWE-bench and halves costs suggests an initial technical wedge [LINUX DO, 2026].

Data Accuracy: YELLOW -- Competitive analysis is inferred from product positioning and market segments; no direct competitor data is publicly cited.

Sources

1. [LithosAI.com, 2026] Home | LithosAI | https://www.lithosai.com/
2. [GitHub, 2026] GitHub - lithos-ai/motus | https://github.com/lithos-ai/motus
3. [SOTA Sync, 2026] Motus：一条命令起 Agent 服务，开源版「Agent 部署平台」 | SOTA Sync | https://sotasync.com/reader/2026-04-15-motus-open-source-agent-serving/
4. [CMU SCS News, 2026] SCS Team Wins Meta Award for Work To Lower Financial, Environmental Costs of AI | https://www.cs.cmu.edu/news/2022/ai4ai-meta-award
5. [CMU SCS, 2026] Zhihao Jia - CMU School of Computer Science | https://www.cs.cmu.edu/~zhihaoj2/
6. [LINUX DO, 2026] CMU教授开源Agent框架Motus，多模型编排SWE-bench跑到79%且成本减半 | https://linux.do/t/topic/1974190
7. [Perplexity Sonar Pro Brief, 2026] LithosAI Overview