Mapped advisor directory groups to FinDesk workspaces and agent scopes. Two legacy groups don’t map cleanly; IAM is deciding whether to retire them.
Next: decision from IAM before scope freeze
Services · Forward Deployed Engineering
Between a working demo and governed production sits the groundwork: identity, data access, your ontology and knowledge graph, policy, audit, model access and the people who will run it. Our Forward Deployed Engineers do that work inside your environment, alongside your teams, and then hand it over. FinAI OS, the managed agent platform, runs as cloud-native containers, on-prem or in your private cloud.
Schematic, not to scale. Blockers and their order vary by institution.
What ‘groundwork’ means
We work through each with your teams. For every area, the right-hand column is what stays with you when we leave.
| Area | FinAI OS | What we do | What you keep |
|---|---|---|---|
| 01Environment & perimeter | All layers | Map where models, data and agents will run; agree network boundaries and the container platform with your infrastructure team. | A reference deployment architecture for your infrastructure |
| 02Identity & entitlements | FinDeskFinHubFinClawFinGuard | Connect your identity provider; map roles and entitlements to workspaces, tenants and agent scopes. | Role mappings your IAM team owns and maintains |
| 03Data & knowledge | FinLore | Connect your data sources, then clean and resolve the data into an ontology and knowledge graph of your business, in your own environment, by design. Our engineers use agents to help with the work, and your experts sign off. Set up knowledge bases that agents can cite, scoped to who may see them. | An ontology and knowledge graph in your own environment, by design, with named owners, knowledge bases agents cite, and the rules for what each agent can read |
| 04Policy & controls | FinSAFEFinGuard | Start FinSAFE in audit mode to record what agents do, then draft policies with compliance: what agents may do alone, what needs approval, what is never allowed. | Versioned policies your teams can review and change |
| 05Sandboxing | FinSAFE | Isolate tool and code execution; agree what each agent can reach; prove the boundaries hold. | Execution boundaries that are documented and that you can re-test |
| 06Agent runtime | FinHubFinClaw | Stand up the agent platform and tenant isolation on your infrastructure; connect tools to your systems. | A runtime your platform team operates |
| 07Surfaces | FinDesk | Brand the desktop; configure workspaces and plugins for each role. | A branded desktop, plus the plugin SDK for your own tools |
| 08Evaluation & monitoring | Cross-cutting | Build evaluation sets with your subject experts; test how agents fail; connect agent telemetry to your monitoring. | Evaluation sets you can re-run on every change, and alerts in the tools you already use |
| 09People & runbooks | Cross-cutting | Train operators, advisors and supervisors; write runbooks with the people who will use them. | Runbooks, trained owners and an escalation path |
Fintology, in design, is our product for the rules, schema and ontology of enterprise knowledge. Today our engineers build your ontology with you.
The engagement
Four phases. Each ends in a gate your institution signs off, and we move on only when it’s met.
Know the environment
Lay the foundations
First workload, for real
Your team runs it
A week in the field
A sample engagement log and board from the Ground phase. Most of the work is unglamorous: mappings, policies, findings and fixes, done with your people in the room.
Mapped advisor directory groups to FinDesk workspaces and agent scopes. Two legacy groups don’t map cleanly; IAM is deciding whether to retire them.
Next: decision from IAM before scope freeze
Drafted FinGuard policy WM-ADV-07: product recommendations are held for licensed advisor review; cash holds and fact-finding are allowed. Compliance edits accepted into v0.3.
Finding RT-014: the agent could be led into paraphrasing a draft house view not yet approved for clients. Fix: FinLore scope limited to published documents; regression case added.
Status: fixed, re-tested, closed
Ran the advice-scenario evaluation set against the agent and policy v0.4. Flagged cases reviewed line by line with two senior advisors; two rules tightened, one test case rewritten as ambiguous.
Mapped households to accounts from the portfolio system for FinLore and the agent’s tools. Reconciliation shows joint accounts double-counted; fixed at the source mapping, not in the agent.
Platform team ran the staging deploy runbook unaided. Two gaps found in the rollback steps; runbook updated by the platform team, not by us.
Engagement models
Scope is agreed after the assessment. These are starting shapes, not packages.
Starting small? The 8-week visibility pilot is the lowest-commitment way in. →
| Compare | Model AGroundwork firstGet the environment ready before choosing a workload | Model BSolution deploymentOne solution, such as Personal Wealth Agents, taken to production | Model CPlatform rolloutFinAI OS as a shared platform for many teams |
|---|---|---|---|
| Best for | Institutions with demos but no production path | A business line with a clear first use case | A platform team serving several business lines |
| Phases | Assess · Ground | Assess · Ground · Deploy · Hand over | All four, repeated per workload |
| In scope |
|
|
|
| Typically involves | FinGuardFinSAFEFinLore | FinDeskFinHubFinClawFinGuardFinSAFEFinLore | The FinAI OS products your teams need, shared across business lines |
| You end with | A governed environment ready for a first workload | A supervised workload in production, run by your team | A shared managed agent platform your platform team runs |
The 8-week visibility pilot
Before you block anything, find out what’s happening. For eight weeks, FinSAFE’s desktop build runs in audit mode on one department’s desktops, recording what agents do before any policy is tightened. Nobody changes how they work.
Run by our Forward Deployed Engineering team. After the pilot, a natural next step is to switch the policy on and bring agents onto the platform.
What you’ll need to decide: a notice to employees about the endpoint sandbox, and how long audit data is kept and masked.
Start with the 8-week pilotHandover principles
The goal of an FDE engagement is that you need us less at the end of it.
Everything we build runs in your environment, as cloud-native containers, on-prem or in your private cloud, under your change control. We work through your processes rather than around them.
Credentials, policies and configuration belong to your teams. We work under your access controls, and you can revoke our access when we’re done.
Runbooks are written with, and then by, the people who will use them. Your platform team runs the final deploy unaided before we step back.
At handover you get a written list of open risks, known limits and recommended next steps, including the ones that aren’t ours to fix.
Product support continues under your FinAI OS agreement. Further FDE time is optional, not assumed.
Talk to FDE
Tell us what you’re trying to deploy and where. An engineer, not a sales script, will reply.
Next step
Talk with a Forward Deployed Engineer about FinAI OS, a single product or the wealth solution. Or start by seeing what your agents already do, with the 8-week visibility pilot.