The agent layer for
transaction services.
Agents that read the data room, build the analysis and draft the report. Your team reviews and signs.
A new way to run a diligence.
Agents change how the work gets done, not who does it. The same way of working holds whatever the size of the team.
Agents carry the tie-outs, the chases and the exhibits, so every workstream arrives at review already evidenced.
The same scope, the same audit trail, the same standard of report, run end to end by the people who signed it.
What sits under a report.
Models at the bottom, your signature at the top. Each layer does one thing, and each one can be checked.
- 01
Models
Frontier models from Claude, OpenAI or Grok, on your own subscription. Swappable, never a lock-in.
- 02
Harness
Planning, tools and tests around each model call. One agent, one job, one output, checked before it lands.
- 03
Your data
The data room, the trial balance, the databook. It stays in your environment and never trains anything.
- 04
Your method
Your report template, your databook conventions, a few past reports. Encoded once, followed on every deal.
- 05
The agents
Families of agents: data room, earnings, balance sheet, report. Each one writes to your method.
- 06
Your review
Scope, open items, every number traced to its source cell. Judgement and sign-off stay with your team.
The databook is built before you open it.
One mandate, start to finish. You start at the judgement calls.
The app is the easy part.
Set the scope, then work the list. It is built to the way your team already runs an engagement, on the model and the files you already have.
Scope
Name the entities, the periods and the workstreams once. Every agent works to that scope, so a tie-out in Germany and a peg in the UK land in the same engagement.
Open items
Every gap an agent finds becomes an item with an owner and a status. Client answers land against the item that raised them, and the completeness log updates itself.
Built to your method
Your report structure, your databook conventions, your order of priority. Nothing new for the team to learn.
Plug in the model you pay for
Claude, GPT, Grok, or a model on your own machine. Switch mid-engagement, nothing else changes.
No migration, no new file store
Workbooks, decks, documents and PDFs, read from your folders and written back in the same formats.
Four steps, in the order you already work.
You keep your method. The agents sit inside the process your team already runs and take the assembly work.
-
01 / INGEST
Read the room
Point an agent at the data room. It indexes every workbook, ties the trial balance to the databook, and lists what is missing.
-
02 / ANALYSE
Build the analysis
EBITDA bridge, normalised working capital, net debt and debt-like items. Each figure carries its source cell. Each adjustment carries its evidence.
-
03 / DRAFT
Write the sections
Sections come back in your template and your house voice. Findings are ranked by what moves price.
-
04 / COORDINATE
Keep the client moving
Open items become an information request list. Client answers land against the item that raised them, and the completeness log updates.
Dozens of agents, grouped into families.
A diligence report is a few hundred small jobs. Each agent does one of them: a tie-out, a peg, an exhibit, a chase email.
- data room indexerVDR export · 412 files → index · 9 folders412 files, 3 duplicates58 s
- file classifierindex → file map398 classified, 14 unknown22 s
- trial balance tie-outTB_FY22-24.xlsx · GL extract → databook · tab 03411 lines tied, 1 exception41 s
- databook builderTB · mapping → databook_v4.xlsx6 tabs, 23 linked cells37 s
- completeness logindex · scope → gaps.md9 gaps open9 s
- IRL managergaps · client replies → IRL · 14 items6 answered, 8 chased12 s
A databook that ties to the trial balance, and a written list of what is still missing.
- EBITDA bridge builderdatabook · tab 03 → bridge_FY24.xlsx23 adjustments, 7 lines41 s
- add-back adjudicatorbridge · your policy → adjudication log17 accepted, 6 held for you64 s
- run-rate and annualisationmonthly P&L → run-rate tabLTM Nov-24, 2 one-offs18 s
- margin walkrevenue · COGS by SKU → margin walk tabFY22 to FY24, 4 drivers33 s
- cohort analysisinvoices · customer list → cohorts tab12 cohorts, NRR 108%47 s
- revenue by customerinvoice register → top 20 tabtop 20 = 61% of revenue15 s
- related-party screenGL · entity list → RP note3 transactions flagged29 s
The bridge from reported to adjusted EBITDA, with every line tied back to a source cell.
- NWC normalisationmonthly balance sheets → NWC tab24 months, 4 reclasses52 s
- seasonality and pegNWC tab → peg memopeg 4.1m, 12-month average26 s
- debt-like classifierBS · contracts → net debt tab9 items, 2 held for you38 s
- restricted cash screenbank statements → cash note0.6m restricted14 s
- capex and lease reviewFA register · leases → capex tabmaintenance vs growth split31 s
- EBITDA to cashbridge · NWC · capex → cash conversion tabconversion 74%11 s
A peg with twelve months of seasonality behind it, and net debt with each debt-like item named.
- findings rankerall tabs · logs → findings.md18 findings, ranked by price impact21 s
- section drafterfindings · your template → report · section 4draft 4.2, in your voice73 s
- exhibit builderdatabook → 41 exhibits41 exhibits, your format56 s
- cross-reference checkerreport · databook → check log212 references, 0 broken19 s
- management call prepheld items · gaps → call sheet11 questions, ordered8 s
- audit trailevery run → trail.jsonevery figure to its cell3 s
Your report, in your template, in your order of priority. Every number stays traceable to its source.
One mandate fans out into families of agents. You describe the engagement the way you always have; the split into jobs is ours.
Client data does not leave your environment.
Diligence material is the most confidential data a firm handles. The product is built around that.
Beta firms get the data-handling note, the model setup and a walkthrough of what is logged, before any deal data is loaded.
- Runs locally or in your own tenantnothing is uploaded to us
- Bring your own model subscriptionno tokens resold, no quota caps
- No client data used for trainingwritten into the contract
- Every figure traces to a source celldatabook reference on every line
- No output leaves without a reviewera person signs, not the model
- Built by people who ran the worktransaction services, M&A, investment
Why we built it
We did this job for years. Building the databook took days. Reading it took a morning.
So we built agents to do the building. They follow your method, they never make up a number, and the person who signs still decides.
What partners ask first.
01Does this replace the analyst?
No. It takes the reformatting, the tie-outs, the file chasing and the rebuilt schedules. Judgement, client calls and sign-off stay with the team.
02Where does our client data go?
Into your environment and nowhere else. The agents run locally or in your own tenant, on your own model subscription. We do not hold deal data and we do not train on it.
03How do we know a number is right?
You can open it. Every figure carries its source cell and every adjustment carries its document. Anything that rests on a judgement call is flagged.
04How granular does it get?
Down to the artefact. There is no single diligence agent. There is a tie-out agent, a peg agent, an exhibit builder, a chase drafter. Each has one job, one output and its own tests.
05Will it work with our report template?
Yes. We take your report structure, your databook conventions and a few past reports, and the drafting agents write to them.
06What does the private beta involve?
One live engagement, run in parallel with your normal process, so you can compare the output. We take a few firms at a time and work with the team on the first deal.
07What does it cost?
Nothing during the beta, and no commitment beyond an NDA. Pricing comes after, and you will see it before anyone asks you to pay.
08What is it bad at?
Anything that needs a room. It will not read a management team, negotiate a peg or decide how hard to push an add-back. It also needs a real data room. Point it at forty scanned PDFs and it will tell you the room is incomplete.
Bring the agents to your next engagement.
Private beta, a few firms at a time. Free while it runs, and no client data needed to evaluate it.
We open the beta to a few firms at a time, and you will hear from us when a spot opens. A confirmation is on its way to your inbox. If you have a live deal coming up, reply to it and tell us: firms with a live engagement go first.