What Kuzco does

Kuzco (operating brand: Inference.net) is a developer platform founded in March 2024 by Sam Hogan and Mike Pollard in San Francisco, CA, focused on building, monitoring, and continuously improving AI agents in production. The platform enables engineering teams to automatically collect execution traces from their agents, curate training data and evaluations, and train and deploy specialized models targeting frontier quality (claimed to reach Opus 4.6 quality). Core products include Catalyst (the primary training and deployment platform), HALO (Hierarchal Agent Loop Optimizer, an RLM-based technique that recursively self-improves agents by analyzing execution traces and recommending code changes, available as a free open-source desktop application), Schematron (an HTML-to-JSON model delivering Sonnet 4.5 quality at 9B model prices, processing 4M requests per day with 10% week-over-week volume growth), and Agent Signals (a real-time classification and alerting system for production agent issues).

The platform architecture is built on native OpenTelemetry-compatible tracing and integrates with established observability tools including Langfuse and Arize. The technical approach emphasizes cost efficiency: demonstrated benchmarks show a 1B specialized model replacing a 70B frontier model with comparable quality, 5.7x lower p99 latency, 10x lower inference cost, and 70x smaller model size. The company is SOC 2 Type 2 compliant, positioning it for enterprise customers with data residency and privacy requirements.

The business model is multi-tiered: freemium (HALO Desktop, free and open-source; $50 Schematron credit for new users), usage-based API access (Schematron and model inference priced per token/request), and quote-based enterprise annual subscriptions (Catalyst platform for teams shipping agents in production). The go-to-market motion is product-led growth, with developer relations and community sponsorships (e.g., Tiny Models DPO video) supplementing direct enterprise sales. Total disclosed funding is $22.5M: $22M from Compliant VC across two $11M rounds in 2024, plus $500K from a16z crypto in September 2024 (the latter signaling strategic alignment with decentralized/distributed LLM inference marketplace positioning). The company is venture-backed and private, with 1-10 employees and named customers including Olive (Food/Consumer) and Gravity Ads (Advertising/Marketing).

Name

Kuzco

Legal name

Inference

Website

https://kuzco.xyz

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

1–10 employees

Short description

Kuzco (Inference.net) is a developer platform for building, monitoring, and continuously improving AI agents in production, using proprietary HALO trace-analysis optimization and a Catalyst training pipeline that trains specialized models from application-generated data to deliver frontier-model quality at lower cost.

Ownership category

akta.pro rank

Product category

AI Inference Platform

NAICS

Software Publishers (5132), Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (518210), Custom Computer Programming Services (541511)

SIC

Services-Prepackaged Software (7372), Services-Computer Programming Services (7371)

akta.pro primary industry

On-Device Inference Runtimes & SDKs (mobile/embedded) (HDAAAJAB)

akta.pro secondary industry

Container Platforms & Orchestration (Kubernetes) (HDABADAF)

Keywords

• Distributed LLM inference
• AI agent infrastructure
• Specialized model training
• LLM inference marketplace
• AI observability platform

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Infrastructure, Technology or R&D, Operations, Marketing or Sales, Supply Chain

Revenue model

1. API Access (Schematron and model inference): Usage-based API access for model inference, including Schematron HTML-to-JSON model and custom trained models. Pricing based on token volume and requests processed.
2. Catalyst Platform Subscription: Enterprise-grade platform subscription for training and deploying specialized AI models with trace collection, data curation, and eval capabilities. Targets teams operating AI products at scale.
3. Developer Tools (HALO): HALO desktop application is free and open source, providing bug finding for agents. Links to code repositories for automated fix recommendations.

Pricing tiers

Go-to-market motion1 record

Distribution channels3 records

Marketing channels4 records

Core offering

Kuzco operates a marketplace for a distributed LLM inference network, with the underlying platform (branded as Inference.net) providing developer tools for building, monitoring, training, and deploying self-improving AI agents. Core sellable offerings are the Catalyst training and deployment platform, the HALO open-source agent optimization tool (including HALO Desktop), and the Schematron inference model for HTML-to-JSON conversion. Revenue is generated through API usage fees, Catalyst subscriptions, and enterprise deployment contracts.

Product overview

Inference.net is a developer platform for building and monitoring self-improving AI agents. The product suite includes Catalyst (core platform for training and deploying LLMs), HALO (agent optimization technique with desktop application), Schematron (specialized HTML-to-JSON model), and Agent Signals (quality monitoring). The platform enables teams to collect traces from production agents, curate training data, train specialized models, and continuously optimize agent performance through automated analysis and feedback loops.

Differentiator

Problem solved

Functional benefit

Brands

• Catalyst: Developer platform for teams shipping agents in production; enables training and deploying frontier LLMs in minutes using application-generated data.
• HALO (Hierarchal Agent Loop Optimizer)
• Schematron
• Agent Signals

Products and services

• Catalyst Enterprise developer platform for teams operating AI products at scale; automatically collects traces from production agents, curates training data and evals, and trains/deploys specialized frontier-quality LLMs using the data the application is already generating.
• HALO (Hierarchal Agent Loop Optimizer) Open-source agent debugging and optimization tool (including the HALO Desktop application) that analyzes agent execution traces and recommends code fixes; 100% free, runs locally, and can import traces from Langfuse or Arize.
• Schematron Specialized HTML-to-JSON inference model for apps and agents, marketed as delivering Sonnet 4.5 quality at 9B model prices with fast inference; available via the inference.net API with $50 in free credit for new users (code EXTRACT, first 100 users).

Quantifiable outcome

• 10x lower inference cost when replacing 70B models with specialized 1B models trained on actual workload
• +3 more outcomes

Named customers2 records

Segments1 record

Ideal customer profiles1 record

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration3 records

AI capability9 records

Feature5 records

Scale indicators4 records

Recent moves6 records

Expansion highlights5 records

Broad incumbents

• Weights & Biases: MLOps platform for experiment tracking, model training, and monitoring. Adjacent in model training and observability workflows relevant to Catalyst and Agent Signals.
• Hugging Face: Broad AI/ML platform for hosting models, datasets, and Spaces, plus Inference API. Comparable on model serving infrastructure and developer community GTM.

Direct peers

• Together AI: Cloud platform for training, fine-tuning, and serving open-source LLMs with a focus on cost-efficient inference. Directly comparable to Inference.net's Catalyst and Schematron offerings in the LLM infrastructure space.
• Fireworks AI: Provides fast, low-cost inference for open-source LLMs and offers fine-tuning capabilities. Highly comparable on inference pricing model and developer-first GTM motion.
• Modal Labs: Serverless compute platform optimized for AI/ML workloads, including model serving and inference. Comparable on developer experience, API-first distribution, and infrastructure focus.
• Replicate: API platform for running ML models in the cloud, with usage-based pricing. Comparable business model (usage-based inference API) and target customer (developers integrating AI into applications).
• Anyscale: Ray-based platform for distributed AI compute, model training, and serving. Comparable in offering scalable AI infrastructure for training and serving specialized models.
• OpenPipe: Platform for fine-tuning LLMs on production data and building specialized agents. Highly comparable to Catalyst's trace-to-trained-model pipeline and agent focus.

Emerging players

• Langfuse: Open-source LLM observability and evaluation platform. Integration partner for HALO Desktop; competes in agent tracing/monitoring space alongside Agent Signals.
• Arize AI: ML observability platform with strong agent/LLM tracing capabilities. Integration partner via HALO Desktop; overlaps on agent monitoring and classification.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights6 records

Customer concentration

• Xx.com/kuzco_xyz

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles2 records

Funding overview

Funding rounds1 record

Investors1 record

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Kuzco