AlphaBitCore

The product exists only because models and agents are being deployed, and it operates on them directly, unifying models, agents, tools and workflows under one governed runtime. The financial services layer is counted rather than described, shipping 23 agents, 95 workflows, 78 skills and 12 models preconfigured for research, portfolio, advisory and compliance work. Its stated purpose is moving AI from pilot to production, which places it squarely inside the deployment problem rather than adjacent to it.

Oversight is the product rather than a feature of it, with policy enforcement in the runtime, approval paths configured to the firm's own structure, and the positioning stated as artificial intelligence that professionals use and compliance approves. Agents operate against a governed, sandboxed surface rather than raw interfaces, which bounds what they can reach. Held at B because no threshold, escalation rule or human approval requirement is specified for particular action classes, so where a person must intervene is left to configuration.

The evidence mechanism is specified precisely and targets the right property. Execution records are sealed and any governed execution can be deterministically replayed from its event stream, which the company summarises as audit moving from log interpretation to execution replay, and the stated capability is that compliance, risk and audit teams can verify exactly what a workflow did rather than only what it produced.

That process-versus-output distinction is the correct one for agentic systems, where the output may be defensible while the path to it was not, and replayability rather than logging is what makes the record testable.

No customer, institution or deployment is named, and no adoption figure or outcome measure is published. The company is seed stage as of early 2026. What is evidenced is the founding team's provenance, drawn from a market data and research firm, a corporate research laboratory, and two capital markets institutions, which is relevant credibility for selling governance to compliance functions but is not evidence the platform is in production anywhere.

No boundary statement was located. The platform is configured against each firm's own policies, data sources and approval paths, which implies tenancy separation without stating it, and a control plane observing complete agent execution across customers accumulates a detailed picture of how regulated firms actually deploy AI. Nothing addresses whether that informs the product or how customer execution records are isolated.

No data protection agreement, retention schedule or subprocessor list was located. The platform sits in the execution path for research, advisory and compliance workflows and retains sealed records of every action, so what those records contain about clients and how long they persist is a question the architecture raises directly and does not answer.

No attestation, certification, trust centre or enumerated control set was located. Sandboxed execution is described as a design property, which is a technical control rather than an audited one, and a compliance function evaluating a platform that will hold sealed records of all AI activity would require documentation that is not published.

One supervisor is engaged specifically and the requirement is stated rather than gestured at: the broker-dealer regulator's emerging agentic AI oversight expectations, with the operative obligation that firms must reconstruct the full chain of agent activity and not merely final outputs, and the workbench is described as aligned to those 2026 expectations. That is a supervisory expectation not previously recorded elsewhere in this index. Held at B because only one regime is engaged and no other regulator, rule or jurisdiction is mapped.

No fairness testing, bias monitoring or disparity analysis was located. The platform governs how agents execute rather than what they decide, which places bias in the customer's application layer, and advisory and research workflows running on it still produce outcomes that differ between clients. Governance content addresses provability and auditability throughout without reaching fairness at any point.

No guarantee, indemnity or correction process was located. The exposure is institutional, since the firm relies on sealed records and replay to demonstrate control to an examiner, and nothing describes where responsibility sits if a replay proves incomplete, an execution record is challenged, or governed activity turns out not to have been captured as claimed.

Twelve models are counted as shipping with the financial services workbench and not one is identified, so a firm knows how many dependencies it is acquiring without knowing whose they are. For a platform whose entire proposition is provability and audit, leaving the model inventory unnamed is the conspicuous gap, since supervisory expectations on third-party models require exactly that inventory.

The integration approach is architecturally interesting rather than a connector list: instead of exposing raw interfaces to agents, disconnected tools, local scripts and remote data are virtualised into a single governed and sandboxed executable surface, with more than ten protocol connectors and primary financial data integrations shipped preconfigured. Held at B because no individual data provider, portfolio system or core platform is named, so a firm cannot confirm its own stack is covered.

No hosting provider, region, residency commitment or self-hosted option was located. Deployment is described in terms of configuration against the firm's policies and approval paths rather than in terms of where the runtime executes, which matters for a control plane that necessarily sits in the path of every governed AI action.

No pricing, packaging or basis of charge was located. The delivery model is described, with the vendor's field engineers configuring the first deployment before the customer's own platform team extends it, which implies a professional services component whose cost is not indicated alongside the platform licence.

The buyer set is named with unusual specificity for a company this young, spanning banks, insurers, custodians, clearing firms, asset owners, quant firms, internal AI platform teams and multi-line institutions on the control plane side, and asset managers, wealth platforms, registered advisers, turnkey asset management programmes, adviser platforms and research teams on the workbench side. Target roles are named as technology and AI leadership. Held at B because none of that coverage is evidenced by a deployment.