Questions the record is built to answer

Fontana is an integrated platform for AI workflows and agents that runs in your cloud or ours. It brings together a web app, a real-time backend, a workflow engine, an identity plane, a secrets store, observability, and a connector runtime, so approved work runs on the record and you can trace any figure back to the file it came from.

There is no single list price, because cost depends on how you deploy, how much compute you run, and how much model use you make. The Cost center inside the product tracks model spend by user, model, gateway, workload, and agent, and lets you cap it, but that is only the token side of the bill. Book a call and we will walk through the cost model for the deployment you have in mind.

Bring one process and a sample of your data to a demo. We will run it live, map it to your standard, show you the breaks, and leave you with the lineage. The Getting Started guide covers setting up your first workspace once you are in.

Yes. Fontana installs onto your own Ubuntu servers from a registry you control, in your cloud or ours. Each workspace is isolated at the namespace level by default, with its own identity, Vault-held secrets, data stores and default-deny networking, and you can choose a dedicated node pool or a dedicated cluster where a contract or regulator requires no shared control layer.

Skills are modular instruction packages an agent loads when it needs them. Each skill is a folder holding a required SKILL.md plus any supporting files, and they live in Fontana Files namespaces, so you can version them and share them between agents.

Fontana is built for SOC 2-aligned diligence, with the controls documented page by page so your security team can review them against your own deployment. Each workspace is isolated at the namespace level, with its own identity, Vault-held secrets, data stores, encrypted volumes, TLS at the public edge and default-deny networking; the cluster control layer and node kernel are shared by default, which Fontana calls soft multi-tenancy and documents plainly, and you can move to a dedicated node pool or a dedicated cluster. Identity and Kubernetes API activity are written to an append-only ImmuDB ledger per workspace, which is the system of record for audit evidence, with application and workflow producers rolling out by release, while Grafana and Loki are only an operational search plane.

Yes, and one precision matters: SOC 2 is an attestation rather than a certification. Fontana AI Limited holds a SOC 2 Type 1 attestation issued in 2026 with Sensiba and Tempo, covering the design of controls at a point in time. A Type 2 examination, which tests operating effectiveness over a period, has not been completed, and ISO 27001 is still in progress. The report is available under NDA, the SOC 2 control summary is the questionnaire index with the control matrix and Trust Service Criteria map, and the vendor diligence FAQ carries answer copy you can paste into a vendor form. **Scope:** production-class hosts are in scope for evidence, while integration and local developer environments are not.

The Knowledge Graph is Fontana's knowledge engine for the operational context your agents need, such as firm rules, mappings, approval criteria, exception history and process documentation. You organise source documents into namespaces and folders, and Fontana derives one governed graph per namespace with four overlays on it: full-text search, a structural document outline, a typed entity graph and vector similarity. Agents retrieve approved, cited context on each turn, with path citations back to the source file, and each agent searches only the namespaces you assign it. You control the schema too, with your own ontology or one Fontana proposes from your corpus, under strict, log or evolve policies for anything outside it.

You design a workflow on the Flow canvas as a graph of nodes joined by edges, where each node does one step and passes typed data to the next. The workflow engine runs it on the server: nodes whose configuration or inputs changed re-run, unchanged nodes reuse their last output, and independent branches run in parallel. Every build publishes each node's output rows into the workspace through Fontana Files, with that port's columns, audit items and lineage, so Explore, the canvas grids and dashboard tiles all read the same rows. You can press Run yourself, leave auto-build on so a change starts the build, or give a node its own schedule.

Fontana is built for organisations with strict compliance, audit and operational control requirements, and the platform is shaped around investment operations across public and private markets. **Operations** prepare the day from drafts that already point at the source file, **control** reviews the breaks and the missing support, and **leadership** sees the unresolved judgments and the approval trail before a figure is released. If your team aggregates data from administrators, custodians, GPs and data providers, onboards new clients, funds or feeds, or manages exceptions through to a signed outcome, that is the shape Fontana is designed for.

Fontana's co-founders are Alexis Wallace Lavigne, Co-founder and CEO; Jethro MacDonald, Co-founder and CPO; and Zoltan Bourne, Co-founder and CTO. Alexis ran investment operations at Challenger and led global go-to-market for accounting and operations at SimCorp. Jethro was Global Head of AI/ML Product at SmartStream, and previously ran investment operations product management at SimCorp. Zoltan has more than 20 years building financial technology, and was a director at Citadel.

The team is in London and Sydney. Fontana AI Limited is company number 16723185.

Fontana comes from the Latin fons, meaning source. It is why every output traces back to one.

You bring data in three ways, all from nodes on the workflow canvas. **File Input:** upload or drag in CSV, TSV, Excel, JSON, XML and financial formats such as FIX, SWIFT MT940 and OFX, which parse into row datasets in your workspace, while PDF, office documents and images route through document parsing instead. **REST:** an external system POSTs JSON to a REST Input to trigger a run, or a REST API node fetches from an HTTPS API when the run reaches it. **Data Connector:** sync Notion, Postgres, S3 and hundreds of other Airbyte catalog sources, running inside your own workspace with credentials held as workspace secrets in Vault.

Yes. Fontana runs on Bring Your Own Key, so inference goes through gateways your organisation already has contracts with, and 23 are built in, covering unified routers such as OpenRouter and LLM API alongside AWS Bedrock, Azure OpenAI, Microsoft Foundry and Vertex AI. **Your own endpoint:** the Custom and Self-Hosted option takes any base URL that speaks the OpenAI /v1 API, which covers Ollama, LM Studio, vLLM, llama.cpp and similar hosts. Keys are written to Vault only and never returned to the browser, scoped at org, team or person level, and administrators approve which model ids are reachable in production. Nothing runs through a shared Fontana model account, so spend, retention and training policy stay on your contracts.

Both. You can chat with Fontana agents from Slack and from Microsoft Teams, and each external thread is a real Fontana thread, so tools, handoff and audit coverage match Flow chat. Each person links their Slack or Teams identity to their Fontana account first, so their messages run with their own permissions and spending cap, and a channel whitelist decides where an agent may speak. Approvals differ by platform: Slack posts an interactive approval card in the thread, while Teams gives you a link back into Flow to approve there.

No. Your files stay in your environment. Fontana does not use customer data to train foundation models. You choose the models, and approved work can run with no model in the loop.

Approved rules run on their own. An agent proposal waits for a person you name. Rule changes, external messages, and overrides need that same approval before anything leaves the record.

Yes. Once a mapping or a check is approved, the workflow runs deterministically. It keeps running when AI is not in use, and the lineage stays on the record.

On the record. You can trace a figure through each transformation back to the file, feed, or document it came from, including the approval that let it through.

In your cloud or ours. Each workspace gets its own namespace with default-deny networking, its own identity, secrets and data stores, and you can move up to a dedicated node pool or a dedicated cluster where your risk model or a regulator requires it. Credentials stay in your vault, and you decide which models and gateways are available.

One process and a sample of your data. We will run it live: map it to your standard, show the breaks, and leave you with the lineage.

Sensiba and Tempo examined Fontana. The SOC 2 Type 1 report, issued in 2026, is available to customers and prospects under NDA.

The ones your organisation already trusts. Fontana fronts 300+ models through the gateways you approve, and a workflow can run with none of them once the rule is signed.