ARQemedes · Agent-Native xP&A
Coming soon
Planning, run by AI agents. Governed by you.
ARQemedes is the agent-native Extended Planning & Analysis platform: specialist AI agents design models, integrate data, budget, forecast, plan scenarios and explain variance — under explicit supervision, with independent review before anything reaches you for approval. Not a spreadsheet. Not legacy EPM with AI bolted on.
Planning, inverted: agents operate, you govern.
Finance teams still treat planning as a quarterly, human-driven scramble across spreadsheets and stitched-together EPM tools. Forecasts go stale the moment they're published, variance explanations arrive late, and workforce, revenue, and expense plans drift out of alignment.
ARQemedes inverts the operating model. Specialist AI agents operate the planning system end to end — model design, data integration and quality, budgeting, forecasting, scenarios and variance — while you declare objectives and approve the moves that matter, instead of driving cells.
Agents reason, recommend, orchestrate and explain. The deterministic Planning Engine calculates — and remains the numeric system of record. An AI response is never an authoritative plan number: every figure is calculated by the engine, reproducible, and traceable from source to approval.
Meet the agents that run your planning
A supervised team of specialists performs the planning work; independent reviewers check it; humans approve material decisions, production models and published plans.
Supervisor
Decomposes your objective, routes work between agents, caps iterations and budgets, and escalates exceptions to you. Never approves its own work.
Requirements & Discovery
Interviews users and analyzes spreadsheets, documents and screenshots to produce traceable planning requirements.
Planning Architect
Defines scope, methodology, model boundaries, calendars and governance for each planning process.
Data Integration & Quality
Connects ERP, CRM, HRIS and data platforms, reconciles totals, detects anomalies and scores planning readiness. It cannot silently waive a failed quality gate.
Model Builder
Proposes dimensions, measures, formulas, forms and workflows — every formula validated by the Planning Engine before review.
Model Reviewer
Independently checks completeness, grain, formulas, circularity, performance and security — and is never the builder.
Planner
Acts as planning lead: develops methodology, requests assumptions and coordinates the domain specialists.
Domain Specialists
Revenue, workforce, expense, CapEx and cash-flow specialists — each submits structured calculation jobs to the engine, never numbers "from their heads."
Planner Reviewer
Tests assumptions, accuracy, consistency, anomalies, sensitivities and policy alignment before a plan reaches you.
Planning Execution
Runs calendars, submissions, approvals, reminders and monitoring — and triggers controlled reforecasts when actuals demand it.
Policy & Controls
Evaluates segregation of duties, materiality, approval authority, compliance and model risk on every mission. Overrides require an authorized human.
Narrative & Presentation
Drafts evidence-linked commentary and board-ready outputs — every number retrieved from approved plan versions.
Built for the office of the CFO.
ARQemedes serves finance teams at enterprise and upper-mid-market companies — roughly $100M to $5B+ in revenue — that run multi-entity, multi-driver planning and have outgrown spreadsheets or legacy EPM. Delivered as multi-tenant SaaS, with dedicated-tenant and private-cloud options for regulated environments.
- Revenue, workforce, expense, CapEx, cash flow and three-statement planning
- Models built from prompts, spreadsheets, documents and screenshots
- Explained variance and refreshed forecasts — without expanding the team
- Delegated execution without surrendering control
Agents reason. MCP connects. WASM executes. The engine calculates.
Probabilistic reasoning and deterministic computation are architecturally separated: agents formulate the work, a supervisor control plane authorizes it, bounded logic runs in a governed sandbox, and the Planning Engine calculates and stores the authoritative numbers.
Supervisor control plane
Workflow routing, planning-scope enforcement, agent permissions, iteration and cost limits, policy and approval gates, human escalation — with audit and observability across all of it.
Secure WASM execution
Agent tool logic runs as signed modules in a sandboxed WebAssembly runtime — capability policies, memory and timeout governors, and an execution ledger that makes every run reproducible.
Pluggable AI, no lock-in
An LLM gateway routes across OpenAI, Anthropic, Google, Azure and local models — with model allow-lists, data-residency controls, and fallback routing by quality, latency and cost.
MCP-native interoperability
Tools and data sources connect through an MCP gateway and server registry, with tool discovery behind a policy proxy — an open ecosystem, not hard-wired integrations.
Deterministic computation plane
The multidimensional Planning Engine owns aggregation, spreading and allocations, driver-based formulas, currency and time intelligence, and forecast and optimization services.
Governed data & metadata
Actuals, master data, dimensions, hierarchies, measures, drivers and assumptions — versioned, lineage-tracked, evidence-linked, and tenant-isolated.
Governed autonomy, full traceability
Agents operate inside explicit permissions, budgets, limits and escalation rules. High-severity review findings block approval, and every published value traces from source record through assumptions, formulas and calculation jobs to review findings and sign-off — with versioned outputs and rollback. Your data is yours: exportable, and never used to train shared models.
Planning shouldn't be a quarterly scramble.
ARQemedes is coming soon. Join the waitlist to help shape the category and get early access.