LithosAI

Cover Block

PUBLIC

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

PUBLIC

This section provides direct links to LithosAI's primary digital assets.

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

Executive Summary

PUBLIC

LithosAI is building an open-source platform for deploying and improving AI agents using signals extracted directly from production, a technical approach that attempts to address a critical gap in the transition from prototype to reliable application [LithosAI.com, 2026]. The company's core project, Motus, is positioned as a model-agnostic serving layer that promises to let developers orchestrate across any foundational model without vendor lock-in, a significant point of differentiation in a market crowded with proprietary, closed frameworks [SOTA Sync, 2026]. This early-stage venture is a pure academic spinout, founded by Carnegie Mellon University professors Dimitrios Skarlatos and Zhihao Jia, whose award-winning 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, suggesting a future commercial strategy likely centered on managed cloud services or enterprise features [LINUX DO, 2026]. No public funding rounds, customer deployments, or significant press coverage have been identified, placing the company in a pre-seed, pre-revenue stage where validation will come from developer adoption of the open-source tool. Over the next 12-18 months, the key signals to watch will be the growth of the Motus GitHub repository, the emergence of a commercial offering, and any initial venture capital backing that would fund a transition from research project to commercial entity.

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

Company Overview

PUBLIC

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, which serves as the initial and primary milestone for the venture [GitHub, 2026].

Details regarding the company's legal incorporation, headquarters location, and exact founding date are not publicly available. There is no record of a formal launch announcement or funding round in major business or technology publications as of this report's publication [Perplexity Sonar Pro Brief, 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, directly 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.

Product and Technology

MIXED

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's central claim is that it can orchestrate across any underlying model, open or closed, while extracting improvement signals from operational traces like task outcomes, latency, and cost [LithosAI.com, 2026]. This positions Motus as a potential infrastructure layer aimed at developers building agents that need to learn and adapt in real-time.

From a technical standpoint, Motus is presented as a developer tool with a focus on simplicity and flexibility. According to a technical overview, it follows a "no-framework principle," meaning developers can use it without modifying existing agent code written for popular SDKs from OpenAI, Anthropic, or Google [SOTA Sync, 2026]. The platform is released under the permissive Apache 2.0 license [LINUX DO, 2026]. Its advertised workflow allows a single command to serve an agent locally for testing and another to deploy it to a managed cloud service, exposing a unified API endpoint in both cases [SOTA Sync, 2026]. The only performance benchmark cited publicly is a claim that 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.

Competitive Landscape

MIXED, 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, if narrow, wedge.

Given the absence of named competitors in the structured facts, the analysis proceeds without a formal comparison table. The competitive map is best understood by segment.

In the broad category of AI agent frameworks, incumbents are dominated by vendor-specific SDKs like OpenAI's Agents SDK, Anthropic's SDK, and Google's ADK. These are not direct competitors but rather potential substrates; Motus explicitly claims compatibility with them, following a "no-framework principle" that requires no changes to existing code [SOTA Sync, 2026]. The more direct challengers are other open-source agent orchestration platforms, such as LangChain's LangGraph and AutoGPT, which focus on the construction and chaining of agents. LithosAI's differentiation rests on the serving and improvement layer, specifically on extracting improvement signals from production traces of already-deployed agents, a focus less prominent in those projects. 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 and lack the open-source, trace-driven learning loop that Motus proposes.

LithosAI's defensible edge today is almost entirely technical and talent-based. The founding team's academic research in systems and machine learning, including award-winning work on reducing the cost of AI for improving ML systems, provides a credible foundation for the core technical challenge of efficient, multi-model orchestration [CMU SCS News, 2026]. The early performance claim,that Motus's multi-model orchestration "achieves 79% on SWE-bench and halves costs",suggests an initial technical wedge on efficiency [LINUX DO, 2026]. This edge is perishable, however. It is predicated on the continued open-source development of a superior orchestration engine, which is vulnerable to being replicated or surpassed by well-funded incumbents or other open-source projects that attract more developer momentum.

The company's primary exposure is its lack of commercial traction and ecosystem lock-in. While its open-source, Apache 2.0-licensed approach encourages adoption, it does not create immediate commercial moats [LINUX DO, 2026]. A competitor like LangChain, with greater brand recognition, a larger community, and established venture backing, could decide to build or acquire similar trace-learning capabilities, quickly neutralizing Motus's differentiation. Furthermore, LithosAI does not own a critical distribution channel or a proprietary dataset; its value is in the software platform alone, which can be forked. Its cloud deployment option hints at a future commercial path, but without announced customers or partnerships, that channel remains unproven.

The most plausible 18-month competitive scenario hinges on developer adoption of the open-source project. If Motus gains significant traction as the preferred way to serve and iteratively improve agents in production, especially among developers wary of vendor lock-in, it could become a standard infrastructure layer. In that case, the "winner" would be LithosAI, as it could use its community to launch a successful commercial cloud service. If, however, adoption remains niche and a major cloud provider or framework (e.g., LangChain) integrates similar functionality into their core offering, the "loser" would be LithosAI, relegated to a research-centric project with limited commercial scope. The verdict in Analyst Notes will likely turn on which of these trajectories materializes first.

Data Accuracy: YELLOW, Competitive analysis is inferred from product positioning and market segments; no direct competitor data is publicly cited. Technical claims are sourced from a single third-party translation [LINUX DO, 2026] and the company's own documentation.

Sources

PUBLIC

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 | Not applicable (aggregated web search results)