Where your data goes, how the agents are trained, where public numbers come from, and what you get back. Five sections, one flow.
When you upload your master file, the data lives in your isolated tenant. Agents read it. Nothing moves out. Nothing is shared cross-tenant. For enterprise customers requiring it, the tenant runs on a customer-keyed dedicated environment.
Each of the 15 specialist agents starts from a corpus of recognised academic, practitioner, and trade sources. Then they sharpen with use — on your training answers, on your team's corrections, on your corporate context.
Every day, three agents publish a concrete question on the user's CV page. The user answers in one to three sentences. The evening reinforcement task distils each answer into a one-line principle, appends it to that agent's best-practices.md, and surfaces it on the public agent CV. Repeat daily. After a quarter, every agent has a living, dated knowledge base built by the buyer who actually uses them.
For external references — price indices, FX, freight, energy, labour cost, sustainability — AkunAI pulls from named, dated, citable sources. Some are public, some are partner-restricted. Every figure on every output is tagged so you can defend it to a CFO.
EIA Brent · ICE Brent · Eurostat HICP energy · ICIS resin (partner) · gas / electricity TSO data.
Eurostat Labour Cost Index · OECD Productivity · national stat-office wage series.
ECB FX reference · Fed H.10 · LIBOR / SOFR / ESTR daily fixings.
EU ETS allowance auctions · CBAM quarterly publications · CDP scoring.
Eurostat Comext · US Census USA Trade Online · OECD International Trade.
Tecnon OrbiChem · Argus · Platts · Fastmarkets — when your tier covers them, clearly marked.
Your panel first. Then verified scrapes (e.g. IATF Oasis registry — honest empty-state until per-CB scraper lands). Apollo / LinkedIn for stakeholder mapping.
Every figure on every output: actual · index · industry-prior · unknown. A buyer who quotes a hallucinated number to a CFO loses the meeting. AkunAI never lets that happen.
The output is a war-room with eleven tabs covering every angle of a senior buyer's pre-meeting prep. Below is the layout from our most-developed demo case (Asiatech Molding, plastic injection, Tier-1 automotive). Same template lands for every case in the demo flow and for any case you build in /build.
Eight outputs a senior buyer walks into a meeting with. Who produces each?
| Capability | Pactum | Keelvar | Coupa | SAP Joule | MS Copilot | ChatGPT Ent. | AkunAI |
|---|---|---|---|---|---|---|---|
| Pre-meeting cheatsheet (sourced numbers) | — | — | partial | — | — | partial | ✓ |
| 3-ply chess plan with branches | — | — | — | — | — | partial | ✓ |
| Defensible-increase calc with provenance tag | — | — | partial | partial | — | — | ✓ |
| Spec-grounded should-price (PFMEA / PPAP) | — | partial | — | partial | — | — | ✓ |
| Value-engineering brief (resin / coating / QC) | — | — | — | — | — | — | ✓ |
| Power map (LinkedIn / Apollo people-graph) | — | — | — | — | — | — | ✓ |
| Voice Coach in the meeting | — | — | — | — | — | — | ✓ |
| Auto-launched RFI / RFQ | ✓ | ✓ | ✓ | ✓ | partial | — | ✓ |
Source: independent analysis based on each vendor's publicly available product documentation, customer announcements, and analyst coverage as of 2026-Q2. This table is a feature-presence comparison drawn from public information. It is not a commercial benchmark and does not assert quality, performance, or fitness-for-purpose for any specific buyer use-case. Capabilities, pricing, and roadmaps may change without notice. AkunAI makes no representation as to the current state of any third-party product. All trademarks belong to their respective owners.
Three live cases. Real-shape data. Pseudonymised. No card.