Spencer Saldana
← all projects

Agents · Multimodal · Enterprise· in progress

Multi-modal Agentforce

An agent that reasons over text, images, and structured data to drive enterprise workflows end-to-end. Built on Agentforce as a reference architecture for the next wave of PortCo automation.

Multi-modal Agentforce
AgentforcePythonOpenAICLIPWhisper

What it's for

Single-modality agents are a useful starting point. Real enterprise workflows are not single-modality. A claims process touches PDFs, photos, voice notes, structured CRM data, and the unstructured chatter sitting in an email thread. An agent that can move across all of those signals is the actually interesting unit of automation.

The hard parts

Routing the right modality to the right tool. Knowing when to ask a human. Keeping the trace of "why did the agent do that" useful for an auditor six months later. None of these are model problems. They're product and architecture problems, and they're the reason most of these systems quietly fail to make it past the pilot.

Status

Active development. Live demo coming.