◆ Assistive Specialist
Takes the client's goals, balance sheet and constraints and builds a full plan (retirement projections, funding gaps, Monte Carlo on the outcomes), then explains the trade-offs in plain language. Emits a stress-tested plan fed to the advisor agent and the suitability gate.
Memory
Working The client's goals, balance sheet and the plan being built.
Episodic Prior plan versions and the client's stated priorities.
Semantic Planning assumptions, tax brackets, contribution limits, decumulation rules.
Procedural Plan-structuring playbooks per life stage and goal mix.
Store File-based memory tool + assumptions warehouse
Orchestration
pipeline MCP
Harness · Managed Agents … session with sandboxed projection code; compaction on long plans.
Tools
›_ Planning + projection engine Code exec { } Account aggregation (held-away assets) API ⌕ Tax + benefits rules (RAG) Retrieval ⌘ Document generation MCP
Evals & guardrails
- Projection assumptions validated against the firm's sanctioned capital-market assumptions.
- Plan recommendations gated by the suitability-oversight agent before presentation.
- The agent produces the plan and analysis; the advisor agent makes the recommendation, the oversight agent gates it.
Frontier edge
- ▲World-model simulation: runs counterfactual life-paths (early retirement, market crash at 62, long-term-care shock) and shows how each bends the funding curve.
- ▲Causal reasoning: attributes a funding gap to the actual driver (savings rate vs. sequence-of-returns) rather than correlating it with a single bad year.
- ▲Continual learning: refreshes the plan automatically when held-away data changes, consolidating which glidepath options the advisor agents actually adopt.
A sample run
Trigger Client, 54, wants to retire at 60 with a known spending target.
- 1Aggregate held and held-away assets into one balance sheet.
- 2Project funding under base, stress and longevity scenarios (Monte Carlo).
- 3Identify the funding gap and model two contribution / glidepath options.
- 4Draft the plan with trade-offs explained in plain language.
Output A stress-tested plan with a 78% base-case success probability and two remediation options, handed to the advisor agent and the suitability gate.
In numbers
6 min
Median plan build
automatic
Plans refreshed on data change