Product · Agent layer
FinHub
The managed agent platform
An agent on its own is not something you can hand real work to. Someone has to decide who may use it, which skills and tools it gets, which models it calls, where it runs and what record it leaves. FinHub does that work. It makes the agents it runs an enterprise resource your teams can assign, constrain, audit and replace, and gives FinDesk's local agents their models and reviewed skills. It manages FinClaw, supports Hermes, and is designed to take other agent runtimes through one contract.
Placeholder for this product’s hero visual (the findesk page uses a full app window). Design pending review.
Where it sits
The platform at the centre of FinAI OS
FinClaw is where agents run; FinHub is what turns them into a managed service. Server runtimes run confined by default in FinSAFE's sandbox service, one sandbox per execution. The platform hosts, governs and audits those server agents, and its shared inference service serves their models. FinDesk connects to the platform's AG-UI gateway, uses the models it manages, and pulls reviewed skills from its catalog into FinDesk's Agent Store; FinDesk's own local agents run sandboxed on the device, and their network connections are logged in FinDesk's on-device Audit center. Skills are checked here before anyone can install them. When FinGuard, the action gateway, holds the credentials to your enterprise systems, it becomes the agents' route to them; the platform is designed to connect to FinGuard directly.
Sold on its own or as the centre of FinAI OS.
Capabilities
What FinHub does
Five things a platform team needs before agents can be treated as an enterprise resource rather than a demo.
- 1
Runtimes you choose, run confined
A fleet manager starts, routes and restarts runtime instances: one shared runtime, one per tenant, or a mix set by tenant policy. It manages FinClaw and supports Hermes through an adapter. Server runtimes run in FinSAFE's sandbox service by default.
- 2
Identity, tenancy and models
Authentication, sessions and tenant-aware identity, with admin roles for the platform, each tenant and reviewers. The model catalogue lives in your own database, so you can switch or mix providers, including models you host yourself, without touching governance or audit. A shared inference service serves the platform's agents, and FinDesk can use the same models.
- 3
Templates, skills and tools, governed together
Administrators decide in one console which agent templates people can use, which skills they can install and which tools agents can reach. An agent's real reach is where those three meet, so loosening one can't quietly undo the others. Every skill is checked before anyone can install it.
- 4
One record of every conversation
Every conversation with the platform's agents is appended to a write-once store. The database refuses edits and deletes, and summaries and memory are derived from that record, so what your auditors read is what was written.
- 5
Streaming and delivery to every surface
An AG-UI gateway streams structured agent events to FinDesk, browsers and rich clients, alongside connectors for enterprise messaging channels. Delivery, inbox and push carry agent messages to clients and staff in the channel they use.
Governed skills
Your teams' scripts become reviewed, revocable enterprise skills
Agents install skills the way people once installed browser extensions: scripts with permissions that nobody reviewed. In FinHub, every skill is checked before anyone can install it. Layered automated security scans and your risk policy decide whether it's approved, sent to a reviewer, or rejected.
- Checked before install. Findings are grouped using the OWASP Agentic Skills Top 10. Some findings, such as hard-coded secrets or path traversal, reject a skill outright whatever its score.
- Reviewed by people. Reviewers approve, reject or disable each skill, and every decision is recorded.
- FinSkills Hub, FinHub's skill catalog: partial-permission approval, quarantine, rollback and version diff when you upgrade.
- Runs where agents run. Approved skills run inside the same FinSAFE sandbox as the agent. Skill actions on your enterprise systems are designed to go through FinGuard.
Standalone · better with
Useful alone. Stronger in the stack.
FinHub gives agents a governed home on its own. Paired with the rest of FinAI OS, the sandbox, the action gateway, the desktop and the knowledge base connect to the same hub.
FinHub alone
FinHub on its own
- Manages FinClaw and supports Hermes; designed to take other runtimes through one contract
- Server runtimes confined in FinSAFE's sandbox service by default
- Every skill checked before install, with human review
- Templates, skills and tools governed in one console
- Model catalogue in your own database
- Write-once conversation record, with delivery, inbox and push
- Cloud-native containers, on-prem or in your private cloud
Better with
Tenant-scoped execution, per-user quotas, scheduled agents and long-running work, behind the platform's runtime contract.
Server runtimes run confined by default in FinSAFE's sandbox service, one sandbox per execution, and tenant policy set in the platform reaches every cell.
FinGuard sits in front of your enterprise systems, holds their credentials and decides whether each action should happen. The platform is designed to connect every agent action, including skill actions, to it directly.
In the wealth solution
The conversation behind every advisor
In Personal Wealth Agents, FinHub is the advisor agent's platform: conversation, memory and tools.
Audit trail and advisor review
Every conversation and advice step leaves an audit trail inside your own environment. Personalized advice and higher-risk cases go to a licensed advisor before the client sees them, and your compliance team sets the review rules.
Conversational KYC (mockup)
Clients can update profile facts in conversation. KYC links to holdings, risk and goals. Suitability judgments come from the institution's own local judgment models and rules engines, working from fixed, system-generated data — not from this product.
Messages that reach the client
Delivery, inbox and push carry the advisor agent's messages. When a message is sent at a moment of truth is set by FinClaw's scheduled agents (see the FinClaw page).
Client app and licensed advisor console shown on the solution page.
Deploy with FDE
A managed agent platform inside your environment
Forward Deployed Engineers stand up FinHub with your identity, data and channel teams, and leave you able to run it.
Talk to FDE- Step 1
Assess
Map your users, channels, models and data stores, and size the platform to your concurrency.
- Step 2
Ground
Deploy the platform, set up identity and tenants, attach FinClaw or another supported runtime, set skill review rules, and turn on the channels your people use.
- Step 3
Hand over
Hand over runbooks so your platform team runs it.
Next step
Turn agents into a resource your organization can govern
Talk with a Forward Deployed Engineer about FinHub on its own or as the centre of FinAI OS.