Radicalbit
Radicalbit is an enterprise AI infrastructure platform for data and platform teams running models in production. It combines MLOps, observability, real-time machine learning pipelines and an LLM gateway, with two components published under Apache 2.0.
What is Radicalbit?
Radicalbit is an enterprise AI infrastructure platform published by Fortitude Group Srl in Milano, Italy, and presented on the site as part of the Fortitude Group product portfolio. Its own summary calls it “the ready-to-use solution for optimizing the management of LLM and ML models in production”, and it is organised as four building blocks rather than as a single application.
The first, MLOps & AI Observability, handles the deployment and serving of machine learning, computer vision and LLM models through either a visual interface or an API. Models can be imported from the MLflow Models API or from a Hugging Face repository, then versioned and compared through A/B testing, multi-armed bandits, shadow and canary releases, with native CI/CD. On the observability side it computes standard metrics — Precision, Recall, Log Loss, F1-Score, MAE, MSE, R-squared, Variance, Correlation — adds Perplexity and Confidence Score for LLMs, and detects drift.
The second, Real-Time Machine Learning, embeds models inside event streaming pipelines. It covers data integrity (outliers, missing values, schema evolution and enforcement), offers a drag-and-drop low-code transformation editor that can be mixed with Python, runs online inference alongside batch and online features in ELT, and can trigger retraining when drift appears.
The third, the AI Gateway, is a single access point between applications and models. It is application-agnostic and supports any model that adheres to the OpenAI standard. It brings advanced routing on latency and cost, exponential backoff retries, multi-model fallback chains, static guardrails (contains, starts with, ends with, regex), PII detection and masking, LLM-as-a-Judge checks, exact and semantic caching with TTL and similarity thresholds, token and rate limiting, and a cost dashboard broken down by route, group and API key. OpenAI, Anthropic, Google Gemini, DeepSeek, Mistral and Azure OpenAI are supported natively, and any OpenAI-compatible endpoint — Ollama, vLLM, OpenRouter or an on-premise deployment — works as well.
The fourth is AI Monitoring, an open source product under the Apache 2.0 licence, with named drift algorithms (Two-Sample Kolmogorov-Smirnov, PSI, Wasserstein Distance, Jensen Shannon Divergence, Chi-Square Test, Kullback Leibler Divergence, Hellinger Distance) and LLM tracing by sessions, traces and spans. The Gateway is published under Apache 2.0 too. Day-to-day platform work revolves around spaces, streams, pipelines, applications, alerts, secrets and integrations, driven from a Pub/Sub API, an Operators API and a Connotation API.
What it does
- Deploy and serve ML, computer vision and LLM models from a visual interface or through an API
- Monitor drift, data quality and model performance once models are running in production
- Centralise and govern traffic towards several LLM providers behind one gateway
- Apply guardrails: regex rules, PII detection and masking, LLM-as-a-Judge semantic checks
- Cap cost and consumption with token limiting, rate limiting, and exact and semantic caching
- Build real-time streaming pipelines with a low-code editor and embed models inside them
- Trace LLM applications through sessions, traces and spans, and alert on custom logic
When to use Radicalbit / When not to
A quick filter to help you decide if Radicalbit is the right fit.
When to use Radicalbit
- Enterprise data and ML teams that take models to production and need deployment, versioning and drift monitoring in one place
- Organisations consuming several LLM providers that want a single control point for routing, guardrails, access and spend
- Teams working on low-latency event streams, who need models embedded directly inside real-time data pipelines
- Companies whose data must stay inside their own perimeter, since an on-premise or private cloud deployment is documented
- DevOps engineers, AI engineers and system administrators who want auditable code and are willing to self-host the two Apache 2.0 editions
When not to use Radicalbit
- Anyone looking for a model or a model provider: Radicalbit sits in front of third-party models, it does not supply one
- Mobile or browser-first users: there is no iOS app, no Android app and no extension, this is server-side infrastructure
- Buyers who must budget without speaking to a salesperson, since no price is published anywhere on the site
- Teams that want to sign up and start on their own: no self-service registration and no online trial were found
- Non-technical end users: even the open source editions require you to host and operate them yourself, with containers, secrets and deployment work
How to use Radicalbit
A typical end-to-end flow, from setup to results.
- Start from the Book a Demo form on the site, which is the only commercial entry point: no self-service sign-up was found
- Alternatively, pull the open source editions from GitHub, radicalbit-ai-gateway and radicalbit-ai-monitoring, and host them yourself
- On the platform, create a space, the container that holds your resources
- Inside that space, work with streams, pipelines, applications and models, importing models from the MLflow Models API or from Hugging Face
- For the AI Gateway, build your application first, then create a project for it
- Write a config.yaml declaring the models and the routes you want to expose
- Upload, approve and serve that configuration; the documentation advertises a Quick Start in five minutes
- Point your LLM client at the Gateway URL with an API key, passing project-name/route-name as the model parameter
- Set user classes and permissions; administrators create seats according to the subscription type
- Follow the Colab notebook examples, then rely on support@radicalbit.ai, the Discord channel for the open source monitoring product, or GitHub discussions for the Gateway
Pros & Cons
Pros
- Two of the products, AI Monitoring and the AI Gateway, are published under the Apache 2.0 licence, so the code can be audited and self-hosted, which is uncommon in this segment
- Broad coverage inside one product family, from real-time streaming pipelines to LLM governance and cost control
- OpenAI-standard compatibility: any compatible endpoint works, including a model you run on your own hardware
- Native provider support is named rather than hinted at: OpenAI, Anthropic, Google Gemini, DeepSeek, Mistral, Azure OpenAI
- Drift algorithms and evaluation metrics are listed by name, which makes the technical promises checkable instead of decorative
- Detailed public technical documentation spread over three subdomains, readable without any authentication
- EU-based publisher in Milano, an explicit position on digital sovereignty and portability, no vendor lock-in claimed, and on-premise or private cloud deployment available. The vendor also advertises a 92% faster time-to-value and up to a 65% reduction in cycle time, but these are its own claims about its own product and nothing indicates they were independently audited
Cons
- No public pricing at all: no pricing page exists, and no monetary amount appears on any page, in the rendered pages or in the raw HTML
- No self-service sign-up and no online trial: the only way in is a demo form, so evaluation depends on the vendor's availability
- No terms and conditions and no legal notice are published on the site
- The privacy policy and cookie policy are hosted by a third party, iubenda, whose robots.txt closes them to crawlers, so their content could not be verified. This is a documentation gap on our side, not a proven shortcoming of the product
- Nothing is published about data retention, sub-processors, a data processing agreement, training on customer data or a minimum age
- No security certification such as ISO 27001 or SOC 2 is displayed for Radicalbit; the ISO 9001 found on the publisher's side certifies the Fortitude Group, not this tool
- Documentation exists in English only, the site in English and Italian, and the open source editions assume you can host and operate the infrastructure yourself
Pricing & Plans
Two very different routes coexist, and only one of them carries any published figure at all, namely none. Two products are free and permanently available under the Apache 2.0 licence: AI Monitoring (OSS) and the open source edition of the AI Gateway. Both are self-hosted, so the real cost is the infrastructure and the operating effort you put behind them; the site describes the monitoring edition as a “transparent, customizable and free solution”. The commercial offer, by contrast, has no published price. No pricing page exists, an absence confirmed by the site's own page inventory and by its sitemaps, and no monetary amount and no pricing vocabulary appear anywhere on the site, in the rendered pages or in the underlying HTML. The only stated route to the commercial offer is the Book a Demo form. The platform documentation refers to seats granted according to the subscription type, but it names no tier and attaches no figure to any of them. Any organisation that needs to budget will therefore have to obtain a quotation directly from the publisher.
Data, GDPR & hosting
A consolidated view of how Radicalbit handles your data.
GDPR overview
Radicalbit is published by Fortitude Group Srl, established in the European Union at Via Tortona, 4, 20144 Milano, Italy, VAT number IT04268680263. Beyond that establishment, the site makes no GDPR compliance claim of its own. The only two GDPR mentions found describe compliance the customer has to reach: the AI Gateway page offers to “instantly validate GDPR workflows, guardrails, and PII masking”, and the LLM Governance page argues that incomplete logging prevents meeting regulatory requirements such as GDPR or HIPAA. Providing compliance tooling is not the same as declaring compliance. No data protection officer, no Article 27 representative and no privacy contact address appear anywhere on the site, and no ISO 27001 or SOC 2 certification is displayed. The privacy policy that would settle the question is hosted at iubenda, whose robots.txt closes it to crawlers; it was not read.
Who owns the data?
No data ownership clause is published on Radicalbit's own site, and no terms and conditions page exists at all: the site's own page inventory and its sitemaps both confirm that absence. The privacy policy and the cookie policy are hosted by a third party, iubenda, whose robots.txt disallows /privacy-policy/ to every crawler. That document was therefore not consulted and the restriction was not circumvented, so nothing at all can be reported about what it says. The only indication available is architectural rather than contractual: an on-premise or private cloud deployment keeps data inside the customer's own infrastructure, and the AI Gateway is presented as an interception layer between applications and models rather than as a data store.
Reuse rights
The site documents what the tool does with data, not what the publisher would do with it. The AI Gateway intercepts LLM traffic, produces structured logging and end-to-end tracing, and exposes metrics per route, per group and per API key. On the protection side, the product page states that it can “Identify and anonymize Personally Identifiable Information (PII) such as name, email address, phone number, IBAN, etc.” and “Implement data encryption and pseudonymization using mapping to provide external models with additional context.” Whether the publisher itself reuses, shares or trains on customer data is stated nowhere on the site, and the contractual document that would answer it — the privacy policy hosted at iubenda — is closed to crawlers by that provider's robots.txt and was not consulted. Absence of publication is not evidence either way.
Data retention & training
Hosting summary
No hosting country and no hosting region is declared anywhere on the site. What is documented is architectural rather than contractual: Radicalbit can be deployed on-premise or on a private cloud, which places hosting inside the customer's own infrastructure, and the AI Gateway can route to internal or on-premise models just as readily as to external providers. The publisher's blog takes a position on European digital sovereignty and on portability, but that is editorial positioning, not a hosting commitment. Two things should not be confused here: the publisher is established in Italy, while the marketing site itself is served from a United States IP address (35.227.194.51, Google LLC) with SiteGround name servers, which describes the WordPress showcase site and says nothing about where platform customer data would live. The document that would settle the jurisdiction question, the privacy policy, is hosted at iubenda and closed to crawlers by that provider's robots.txt, so it was not consulted. In short: self-hosting is documented, and no hosting jurisdiction is committed to in writing on the site.
Things to keep in mind
Risks and trade-offs to weigh before adopting Radicalbit.
- No public price at all, which makes budgeting impossible without going through a salesperson; the only entry point is a demo form
- No terms and conditions and no legal notice are published on the site, so the contractual frame of the commercial offer is unknown before contact
- The privacy and cookie policies sit with a third party, iubenda, whose robots.txt blocks crawlers, so the conditions under which your data would be processed cannot be verified as things stand. It is worth requesting them in writing
- Nothing is published about retention periods, sub-processors, a data processing agreement, training on customer data or a minimum age
- The percentages used on the use-case pages (70%, 80%, 34%, 30%, 85%, 60%, 75% and 40%) carry no source whatsoever. They are market framing statistics and must not be read as results obtained with this tool
- The 92% faster time-to-value and the up to 65% reduction in cycle time are the vendor's own claims about its own product; treat them as marketing figures until you have measured your own
- Three governance details deserve a question before signing: the site's footer previously named a different legal entity with a different VAT number from the Fortitude Group Srl that owns and operates the product today; the LinkedIn link in the footer redirects to a showcase page rather than a standalone company page; and the ISO 9001 certification visible on the publisher's side belongs to the Fortitude Group, not to Radicalbit
Setup & Integrations
Technical difficulty
Setup targets technical staff: the documentation names DevOps engineers, AI engineers and system administrators. The AI Gateway advertises a Quick Start in five minutes, where you write a config.yaml declaring models and routes, upload and approve it, then point your client at the Gateway URL with an API key. Integration afterwards is straightforward. The real workload sits elsewhere: production deployment, secrets management, access control and troubleshooting each have their own documentation section, and the open source editions leave hosting and operation entirely to you. Pipeline building is gentler, with a drag-and-drop low-code editor and Colab examples.
Deployment
Integrations
Behind Radicalbit
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What is Radicalbit?
Who publishes Radicalbit?
How much does Radicalbit cost?
Is there a free version?
Does Radicalbit offer an API?
Which LLM providers are supported?
Can it be installed on our own infrastructure?
How do you get started?
How do you contact support?
What is known about how personal data is handled?
Should you pick Radicalbit?
Radicalbit is a serious piece of infrastructure, technically detailed and documented in public across three subdomains without any login. Its strength is the combination it assembles inside a single product family: MLOps and model observability, real-time streaming machine learning, and LLM governance through a gateway that sits in front of third-party providers. Two of those blocks, AI Monitoring and the AI Gateway, are released under Apache 2.0, so the code can be read, run and kept in-house, which matters to organisations that treat portability as a requirement. Naming the drift algorithms and the evaluation metrics makes the technical promises checkable rather than decorative. The publisher, Fortitude Group Srl, is established in the European Union at a Milano address, and its stated position on digital sovereignty is consistent with the on-premise and private cloud options it offers.
The main reservation is commercial and documentary opacity. No price appears anywhere on the site, there is no pricing page, no terms and conditions and no legal notice, and the only route to the commercial offer is a demo form. The privacy and cookie policies are hosted at iubenda and closed to crawlers by that provider's robots.txt, so their content could not be verified: an inaccessible document, not a missing one. Nothing is published on retention, sub-processors, a data processing agreement or training on customer data. The figures the vendor puts forward, a 92% faster time-to-value and up to a 65% reduction in cycle time, are its own claims and carry no audit; the percentages used to set the scene on the use-case pages carry no source at all and are market framing, not results.
Radicalbit suits an organisation that already has a technical team, a genuine need for centralised AI governance, and the appetite to negotiate terms directly.
- Choosing a selection results in a full page refresh.
- Opens in a new window.