CoModel AI gives general-purpose LLMs the company-specific intelligence they need to reason, simulate, decide, and act safely.
LLMs know language.
They do not know your operating structure.
They do not understand your KPIs, constraints, trade-offs, business rules, decision paths, or internal logic.
That is why generic AI can answer questions, but still struggle to make enterprise decisions.
Your company already has a model.
It lives across ERP systems, spreadsheets, dashboards, SOPs, meetings, finance models, and individual experience.
The problem is that this intelligence is fragmented.
CoModel AI turns business facts into a company-specific reasoning model.
It maps how your company works, what it optimizes for, what constraints apply, what trade-offs matter, and what actions are allowed.
Your LLM knows the world.
CoModel AI knows your company.
Together, they make enterprise AI usable for real decisions.
CoModel AI transforms fragmented enterprise data into a causal decision model, helping teams simulate trade-offs and identify the best path under real-world constraints.
Moves from forecasting outcomes to identifying what decisions can change them.
Connects relationships to levers, constraints, trade-offs, and downstream effects.
Models multi-variable system influence, not just whether one signal predicts another.
Uses uncertainty to optimize decisions, not only estimate confidence.
CoModel AI starts with the operating facts of the enterprise: metrics, entities, rules, constraints, events, levers, objectives, and relationships.
It identifies how business variables move together, influence each other, and create downstream effects.
It captures what is allowed, what is risky, and what the company is trying to optimize.
Teams can test scenarios before changing capital, policy, process, inventory, staffing, or operating plans.
CoModel AI compares possible actions and surfaces the expected impact, risk, confidence, and trade-offs.
Agents can use CoModel AI as a reasoning layer before they recommend, trigger, or execute actions.
Give AI agents a company-specific reasoning layer before they act.
Ask what would happen if a policy, process, target, or constraint changes.
Turn repeated enterprise decisions into structured, traceable, governed workflows.
Understand likely outcomes before deploying changes into the business.
Move from dashboards and reports to simulation, recommendation, and action.
CoModel AI can pair with local language models so enterprise reasoning stays inside the customer’s environment.
The company model remains the grounding layer.
The language model can change.
The company intelligence stays yours.
Simulate economic policies, public programs, infrastructure decisions, budget trade-offs, and downstream effects before implementation.
Model decisions across capital, risk, liquidity, compliance, portfolio exposure, and return.
Optimize inventory, production, fulfillment, working capital, delivery commitments, and operational trade-offs.
Your company already has intelligence.
CoModel AI makes it computable.
An explicit internal mapping step, not another upload file. Review source units and population standard deviations, edge authorization, lever bounds, approvals and costs. Unknown starter values must be completed; they never become zero or approved automatically.
The config object contains policy_version, series, edges, formulas and lever_rules. Series require actual unit and model_scale (observed population standard deviation); baseline is read and checked by the server. A lever_rule needs metric, action_type, action_unit, bounds or allowed_values, authority, approval_status and matching policy_version. Actions: absolute_delta, relative_delta (fraction), set_and_hold, discrete_option (option_values maps options to final levels). Cost-sensitive planning requires cost_per_unit and a declared cost_unit. Approval dates use server time. Only explicitly authorized direct_model_response edges propagate; association_candidate and total_response do not. Nonlinear local approximations require supported_range. Formula arithmetic supports +, -, *, /; derived-only metrics cannot be tuned.
Optional projection_manifest/source_rows belong alongside config in the JSON envelope. Existing six-file evidence remains the source; no seventh CSV is needed. Empty lever_rules means no actions are permitted.
Zero draws is deterministic, not an uncertainty estimate. Positive draws reuse native relationship sampling with shared realizations across candidates. Native sampling budgets apply.
Copy controls to the editor, then explicitly opt in. The JSON replaces the form request without silently merging objectives. State constraints use final business levels; soft constraints need penalties and probability constraints need explicit thresholds and native draws.
Company Mode describes your business with six small CSV files. Each file answers one simple question, and together they let CoModel AI map how your metrics influence one another. You don't need every file to be full to start — but the more complete they are, the more the model can do. This guide is static reference material; nothing here is AI-generated.
An ordered list of your time periods — weeks, months or quarters. Every other file points back here to say when something happened. Give each period a stable id and a sequence number so the order is never ambiguous.
The objects you measure — the company itself, a product line, a store, a region. Each entity has a stable id and a type, and can point to a parent to build a hierarchy.
A dictionary of every metric you track — revenue, ad spend, churn, headcount. This file defines what each metric means (name, unit, category). It does not hold the numbers themselves — those live in Metric Values.
The actual numbers — one row per metric, per entity, per time period. This is the bulk of your data. Each row references a period (index_id), an entity (entity_id) and a metric (metric_id), and carries the value.
Notable things that happened — a price change, a campaign, a policy decision — tied to a period. Optional context that helps explain shifts in your metrics. It can start header-only.
What you know or want: approved goals (objectives), constraints (limits you must respect), and levers (the metrics you can actually act on). Planning features only switch on once this file holds approved objectives and at least one lever. It can start header-only.
Think of it as a few simple links — each file references the others by stable IDs:
A row in Metric Values is only meaningful because its index_id matches a period in Time Periods, its entity_id matches an object in Entities, and its metric_id matches a definition in the Metric Dictionary. Get those IDs right and everything lines up; misspell one and that number is orphaned.
Say you run a retail business and track things monthly:
With those six files, CoModel AI can learn how ad spend and foot traffic move revenue, and later suggest how to adjust the lever to reach your goal.
Don't start from a blank page. Download a blank template (correct headers with a few rows) or a completed example for every file, then adapt it to your own data before uploading.