The AI brain
that runs your company.
Meeting capture, decisions, tasks, sales planning, procurement, and conversational analytics — one AI-driven platform, every department.
More than a task tool. A working brain.
AI as a first-class actor
CAIB doesn't sit beside your team waiting to be asked. It parses meetings, extracts decisions, generates tasks, validates evidence, and escalates overdue work — with a full audit trail of every AI action.
Evidence over honor system
Tasks aren't done because someone clicked "done." They're done because the assignee submitted proof, the AI evaluated it, and a reviewer approved. The system enforces the difference.
Plugged into your existing systems
CAIB reads your databases, email, calendar, ERP, and files. No data migration, no rip-and-replace. The brain attaches to what you already run.
One brain, every system.
- Postgres / Supabase databases
- ERP systems
- Email inboxes (inbound webhook)
- Calendar
- Spreadsheet ingestion
- Email (transactional + scheduled)
- PDF & Excel exports
- Webhook callbacks
- Audit log
- Claude (default) and OpenAI (fallback)
- Encrypted action tokens (one-click vote/approve via email)
- Cron infrastructure for reminders, escalation, and resolution
Board & Governance
Capture board meetings, extract decisions and tasks automatically, run formal voting with quorum and weighted thresholds, and keep an immutable registry of every decision your company has ever made.
- 01
AI meeting parser with source citations
Paste raw meeting notes; the AI extracts decisions, tasks, attendees, and assignees. Every extraction includes the exact quote from the original notes.
- 02
Formal voting — quorum, weighted thresholds, deadlines
Decisions move through draft → voting → resolved with configurable quorum, optional weighted votes (default 60% threshold), and automatic deadline-based resolution.
- 03
Email voting with secure tokens
Board members vote with one click from their inbox. Single-decision links and a batch landing page for multiple pending decisions, all signed with encrypted 7-day tokens.
- 04
Decision registry
Searchable archive of every decision with filters by status, type (policy / task / budget / operational), date, and initiator.
- 05
Auto-resolution and reminder cron
Decisions auto-resolve when deadlines pass; non-voters get reminders 24 hours before close.
Smart Task Manager
Tasks aren't marked done — they're proven done. AI extracts tasks from meetings, validates submitted evidence against the definition of done, and escalates overdue work without human prompting.
- 01
Evidence-based completion
Every task has a written definition of done. Assignees submit proof via web upload, email reply, or link. The AI scores the evidence's match against the definition with a confidence number, and a reviewer approves or requests changes.
- 02
[TASK-XXXX] email-reply submission
Assignees reply to a task email with the subject tag intact; attachments and the reply body become evidence automatically. 0.98 confidence on subject-tag matches.
- 03
Eight autonomous AI agents
Meeting Parser, Email Matcher, Evidence Validator, Summary Generator, Decision-Task Extractor, Email Decision Parser, Recommendation Engine, plus a base agent class. All actions logged with confidence scores.
- 04
Auto-escalation on overdue
A reminder cron fires at T-7, T-3, T-1, on the due date, and +1 day overdue. Past the configurable grace period, the task escalates and the entire board gets notified.
- 05
Status machine
open → in_progress → submitted → approved (or changes_requested → resubmit). Roles strictly limit who can transition what.
- 06
Audit trail with actor type
Every change records created_by_type / updated_by_type (admin or AI), so the trail shows when the AI moved a task and when a human did.
| Category | Share | QoQ |
|---|---|---|
| Boilers | +12% | |
| Radiators | +4% | |
| Pumps | −2% | |
| Pipes | +1% | |
| Fittings | +0.4% |
Data Whizzard
Ask any question about your business in plain language. Trends, anomalies, root causes, and executive summaries — generated in seconds against your live data, with the safety, audit, and validation an enterprise needs.
- 01
Natural-language SQL with safety enforcement
The AI generates a Postgres query, validates it (no DDL, no DML, mandatory tenant filter), runs under 30s/10K-row caps, and explains the result. Every query logged for audit.
- 02
Temporal analysis — trends, seasonality, forecasts
Linear regression with R² confidence, STL seasonality decomposition, moving averages, exponential-smoothing forecasts.
- 03
Statistical engine — Pareto, HHI, Pearson, z-score
80/20 product analysis, Herfindahl-Hirschman concentration, cross-metric correlations with significance testing, anomaly detection via z-score outliers.
- 04
Diagnostic engine — root cause and driver decomposition
When a metric moves, the AI ranks the contributors by dimension, scores its confidence, and proposes drill-down questions.
- 05
Executive reports auto-generated
Variance reports, period reports, and executive summaries — narrative form with highlights, concerns, and opportunities surfaced from the underlying patterns.
Sales Planning
Day-by-day demand simulation, ABC-weighted forecasting, and out-of-stock-aware coverage analysis. The math behind every reorder decision — auditable, testable, and pulled live from your sales data.
- 01
Deficit analyzer — day-by-day OOS-aware burn
Simulates stock depletion day-by-day with sales capped at available inventory, revealing true coverage gaps that aggregate ratios miss.
- 02
ABC-weighted demand forecasting
Products auto-classified A / B / C; A items get a 1.5-month forward buffer, B 1 month, C half a month. Buffer drives coverage end-date.
- 03
Sales normalization with growth alignment
Missing months decay at 5%/month from the anchor; a configurable growth-target multiplier (e.g., 26% YoY) aligns the baseline with corporate targets.
- 04
Dual coverage metrics
Whole-window OOS-adjusted coverage and forward-buffer fulfillment ratio surfaced together. Color-coded thresholds (red <85% · amber 85–95% · green ≥95% · slate >115% overstock).
- 05
Trend analysis with R² confidence
YoY %, growth-vs-coverage ratio, linear-regression trends scored by confidence (high ≥0.7, medium ≥0.4, low <0.4).
| SKU | ABC | Coverage | Order |
|---|---|---|---|
| PPR-25-EL | A | 96% | 240 |
| PPR-32-T | A | 88% | 160 |
| PPR-20-V | B | 72% | 95 |
| PPR-16-EL | C | 62% | 40 |
| PPR-40-Y | B | 81% | 120 |
Procurement Planner
Forecast demand, generate orders that hit 100% coverage, and pack every truck to the kilogram. Across nine supplier brands, with full audit and one-click email dispatch.
- 01
One-click "Add Order" targeting 100% coverage
Computes the exact quantity to fulfill demand through the planning window, accounting for stock, in-flight orders, ABC buffers, and pack sizes, then creates an editable draft.
- 02
Fill 24t — binary-search truck optimizer
Loads every shipment to a uniform coverage ratio across all SKUs while respecting the truck's weight cap. Three rounding strategies eliminate the "stuck below target" cliff.
- 03
Multi-supplier workflows
Nine active brands run on the same engine, each with their own truck targets, lead times, pack sizes, and email templates. Add a new supplier without touching the math.
- 04
Order lifecycle with full audit trail
Draft → confirmed → sent → received. Inline editing of quantities and dates with instant recalculation. Every change logged with actor (admin or AI). Soft delete preserves history.
- 05
One-click email dispatch with attachments
Configurable To/Cc/Bcc per supplier brand. Excel and PDF order exports auto-generated and attached. Status tracked from send to receipt.
Market Research
Track competitor prices across product categories, match competitor SKUs to yours by spec, and see exactly where you're cheaper, where you're not, and by how much.
- 01
Smart product matching by spec
Competitor products auto-paired to yours by capacity (kW) or other key spec; no manual mapping required.
- 02
Multi-retailer, multi-brand tracking
Eight-plus retailers monitored, fifteen-to-twenty-five brands per category. Country-of-origin captured per product.
- 03
Gap analysis with multi-filter drill-down
Filter by brand, retailer, country, subcategory, model. Sort the comparison table by price, brand, or % gap. Color-coded badges show position at a glance.
- 04
One-click competitor links
Every row links straight to the competitor's product page, kept current with the latest scrape.
- 05
Five product categories out of the box
Boilers, radiators, air conditioners, water heaters, and a catch-all category — extensible to others without code changes.
From boardroom to done.
- Step 01MeetingCaptured and parsed by AI.
- Step 02DecisionVoted on and resolved.
- Step 03TaskCreated and assigned automatically.
- Step 04EvidenceSubmitted and AI-validated.
- Step 05DoneApproved by a reviewer. Audited.
AI watches every step — extracting, reminding, validating, escalating.
Built for the people who run the company.
Executives & board members
One source of truth for decisions, voting, and the work that follows.
Operations & department heads
Tasks that escalate themselves, evidence that proves progress.
Procurement & supply chain
Plans that survive contact with reality, trucks that load themselves.
Sales & commercial
Forecasts grounded in actual sales, ABC tiers that hold up in audit.
Analysts & finance
A 300K-row dataset and twenty algorithms, accessible by typing a sentence.
CTOs & operators of the business
One platform plugged into existing systems, multi-tenant from day one.
Built like the auditor will read it.
Multi-tenant from the database up
Postgres row-level security enforces organization isolation on every table, every query.
AI is read-only on your data
No DDL, no DML, mandatory tenant filter, 30-second timeout, 10K-row cap. Every query logged.
Immutable audit trail
Every action records actor (admin or AI), timestamp, and outcome. Soft-deletes preserve history.
Encrypted email actions
One-click voting and evidence submission use signed tokens with 7-day expiry.
No vendor lock-in on AI
Multi-provider (Claude default, OpenAI fallback). Change providers without changing the product.
Bilingual data layer
UTF-8 throughout; supports Latin and non-Latin alphabets including Georgian.
Modular subscription. Pay for what you use.
CAIB is sold by module. Companies subscribe to the modules they need today and add others as they grow. Final pricing depends on the module mix, data volume, and number of seats. Contact us for a quote tailored to your team.