Researching a major donor or foundation by hand means a dozen browser tabs, a guess at how much to ask for, and no record of why you approached them that way. Prospect Intelligence runs a staged research pipeline against one prospect at a time — discovery, deep background, relationship mapping, mission-fit and capacity scoring, and a cultivation strategy — and hands you a single evidence-backed dossier instead.
Prospect Intelligence is real, working code — not a mockup or a roadmap slide. It has its own database schema, a staff-facing dashboard inside the platform (research monitor, prospect list, agent activity, human review queue), and a working orchestrator that drives a research run through every agent family end to end.
What it isn’t yet: turned on for every organization. Access is gated behind a single organization-level rollout flag that defaults to off — if the flag service is unreachable or unset for your organization, the feature stays hidden rather than opening up by accident. The platform’s own rollout plan (a small named pilot group first, then a wider expansion, then general availability) hasn’t started yet. We’d rather tell you that plainly than show you a screenshot that implies your account can use this right now.
Each step below mirrors a real stage in the orchestrator (src/lib/pil/research-orchestrator.ts).
A research run starts from a goal — a name to look into, or a natural-language request — and is driven by a single orchestrator through its agent families in a fixed order: Supervisory, Discovery, Prospect Intelligence, Relationship Intelligence, Qualification, Strategy, Knowledge Integrity, Operations, and Application. Each agent in those families has a defined mission and (for most of them) an explicit human boundary it cannot cross on its own.
Progress is checkpointed after every family, not just at the end — if a run is interrupted, it resumes from the next incomplete family instead of starting over. If any agent comes back needing a person's judgment, the run pauses in a resumable state and waits for that review to be resolved rather than skipping ahead or failing outright.
The last step before a run completes is always dossier synthesis — a dedicated agent turns everything gathered into a single narrative dossier with facts and inferences explicitly separated, evidence citations, confidence scores, and recommended next actions. Once that's done, qualified prospects can be added to AutoApply's queue, where nothing is submitted until a person approves.
No agent in this pipeline contacts a prospect. Every one of them researches, scores, or drafts a recommendation for a person to act on — approving actual outreach is its own explicit review type, separate from every other kind of check below.
Human review queue. Identity linkage, capacity determinations, policy exceptions, requests for broader agent permissions, high-impact actions, and outreach approval all route to a queue a person has to resolve. An agent that hits one of these doesn’t skip it or fail the run — it pauses in a resumable state until a person decides.
An agent watches the agents. A dedicated governor agent monitors every active run for policy violations and can terminate one immediately — and by design it cannot approve its own authority increases or policy exceptions; only a human decision can resolve those.
Every action is logged. An append-only audit trail records every agent and human action, and a policy engine attaches an allow, deny, or require-human decision to each one — so any recommendation the system produces can be traced back to exactly what it saw and why it decided that.
Tracked in a source registry as permitted, restricted, or prohibited before any agent can query them, with a rate limit and a stored, hashed snapshot behind every claim.
Access is org-scoped end to end — every table carries an organization id, and the rollout flag itself is evaluated per organization, so one team’s access never leaks into another’s. If the flag service is unreachable or a flag is unset, the feature fails closed and stays hidden rather than opening up by accident. Every claim in a dossier is tied to a stored source snapshot with a content hash, so evidence can be checked against what was actually retrieved, not just trusted. Cost budgets can be set at the organization, agent, or single-run level, with an optional hard stop. For the platform-wide security and governance model, see the Trust and Governance page.
Run the free Funding Potential Scan first — a short, no-signup check of the kind of funding your organization is likely to qualify for.
Prospect Intelligence is in controlled rollout, not yet open to every account. Talk to us about early access, or see the governed pipeline it feeds once a prospect is qualified.