Giacomo Nicoli left Boston Consulting Group to build an AI company. He found his technical co-founder, Roberto Stagi, who spent a decade building products at Italian tech giant Bending Spoons [LinkedIn]. Their bet, Ratel, is not another agent framework. It is an open-source context gateway, a piece of infrastructure designed to sit between a developer's code and the large language model, deciding what tools the AI needs to see [Startup.eu, October 2026]. In production, the company claims this approach has cut token usage by up to 95 percent [Founding growth]. For teams running AI agents at scale, that is not an optimization. It is the difference between a prototype and a product.
The Context Gatekeeper
Most AI agent systems today suffer from context overload. Developers give their agents access to dozens of tools, APIs, or skills, and the entire catalog gets dumped into every prompt sent to the LLM. This is expensive and inefficient. Ratel's core proposition is a selective retrieval engine. Its TypeScript and Python SDKs, which are thin bindings over a shared Rust engine, analyze each agent turn and inject only the relevant tools or skills needed for that specific task [AI Engineer]. The engine uses a combination of BM25 and semantic retrieval, requiring no separate vector database [daily.dev].
The product's second function is a self-improvement layer. It analyzes production traces, identifying repeated workflows, failed tool calls, and token-heavy steps. Developers can then test proposed fixes against historical traffic before deploying changes [Startup.eu, October 2026]. This closed-loop debugging is aimed squarely at engineers moving agents from demo to production. The entire stack is open source and runs locally or with OpenTelemetry support, a deliberate wedge for developer adoption [AI Engineer].
The Team and Early Traction
The founding duo brings a classic operator-investor pairing to the Milan tech scene. Nicoli handles strategy, growth, and fundraising, leveraging a background that includes an Entrepreneur-in-Residence stint at lead investor Italian Founders Fund [LinkedIn]. Stagi, as CTO, leads product and engineering, drawing on his experience scaling software at Bending Spoons [LinkedIn]. By June 2026, they had grown the team from two to six, adding an early hire with experience from game development, PayPal, and a Bay Area startup [LinkedIn, June 2026].
Traction, while early, shows cross-Atlantic interest. The company reported onboarding its first customers in Italy and the United States by May 2026 [Founding growth]. More notably, it claimed over 3,000 daily users were interacting with a Ratel-powered agent by that same date [Founding growth]. These are adoption metrics, not revenue figures, but they signal developer activity around the open-source project.
| Metric | Value |
|---|---|
| Pre-Seed Round | 1.1 M USD |
| Team Size (June 2026) | 6 people |
| Reported Daily Users (May 2026) | 3000 users |
The Funding and the Italian Angle
In October 2026, Ratel closed a €890,000 (approximately $1.1 million) pre-seed round [Startup.eu, October 2026]. The lead investor was Italian Founders Fund, with participation from Vento, Italian Angels for Growth (IAG), and Angel Invest. The round is a marker for Italy's emerging deep tech ecosystem, backing a technical founder from a flagship local company (Bending Spoons) and a commercially-minded CEO with ties to the fund itself.
The capital is earmarked for expanding the team and accelerating product development. For investors, the bet is on Ratel becoming the default context layer for production AI agents, a critical piece of infrastructure in a market still being defined.
Where the Wheels Could Come Off
The ambition is clear, but the path is crowded with technical and commercial unknowns. Ratel's early value proposition rests on a few key risks that will determine its next phase.
- Performance claims. The headline 95 percent token reduction is dramatic but comes from the company's own materials [Founding growth]. Independent, third-party validation of these benchmarks in diverse production environments will be crucial for enterprise credibility.
- Commercialization path. The product is open source. The company's stated model is "Open Source / Commercial," but specific plans for monetization are not public. Successfully converting developer adoption into paid enterprise contracts is a classic, difficult pivot for infrastructure startups.
- Competitive landscape. While no direct competitors are named in sources, the problem of agent context management is not obscure. Larger cloud providers and existing AI infrastructure companies could build similar capabilities or acquire competing startups, changing the market dynamics overnight.
The company's answer to these risks appears to be speed and specificity. By focusing narrowly on the context retrieval and debugging layer, and by embedding deeply with SDKs for popular frameworks like Vercel AI SDK and Pydantic AI, Ratel aims to become indispensable to a specific developer workflow before broader competition arrives [AI Socratic].
The Next Twelve Months
For a pre-seed company, the next year is about proving the model beyond the initial wedge. Key milestones will be converting early pilots into announced enterprise customers, publishing detailed case studies that substantiate the performance claims, and likely beginning the process of raising a seed round to scale the commercial effort.
The $1.1 million pre-seed, led by Italian Founders Fund with Vento and IAG participating, gives the team runway to execute [Startup.eu, October 2026]. The question for Nicoli, Stagi, and their backers is whether the developer love for an open-source context gateway can be packaged into a business that owns the plumbing for the next generation of AI agents. In a market obsessed with model size, Ratel is betting the real bottleneck is context. Will developers pay to remove it?
Sources
- [Startup.eu, October 2026] Milan-based Ratel raises pre-seed funding led by Italian Founders Fund to improve AI agents | https://www.startup.eu/investments/ratel-pre-seed-10-2026
- [AI Engineer] Ratel: AI Agent Context Infrastructure | https://ai.engineer/orgs/ratel
- [Founding growth] Founding growth | https://app.notion.com/p/Founding-growth-34bf341af2a380029d45ebd41227a9f1
- [daily.dev] GitHub - ratel-ai/ratel: Context engineering for AI agents. ~80% fewer tokens. | https://daily.dev/posts/github---ratel-ai-ratel-context-engineering-for-ai-agents-80-fewer-tokens-fix-tool-overload-sk-qzjpnumfv
- [LinkedIn, June 2026] Ratel | LinkedIn | https://www.linkedin.com/posts/ratel-ai_ratel-earlystage-hiring-activity-7473357445554225153-3Rv-
- [LinkedIn] Giacomo Nicoli - Ratel | LinkedIn | https://www.linkedin.com/in/giacomo-nicoli/
- [LinkedIn] Roberto Stagi - Ratel | LinkedIn | https://www.linkedin.com/in/robertostagi/
- [AI Socratic] Review: Ratel, context engineering for production agents | AI Socratic | https://aisocratic.com/posts/review-ratel-context-engineering-for-production-agents