Answers that show their working
Node & Nerve links the records your business already keeps into one graph, then answers questions with the entries, links and sources behind every figure, and says what is still missing.
Create a workspace for your company and connect Google Sheets, or explore two fictional companies without signing in.
Your records already describe the business. They just aren’t connected.
A purchase sits in one book, the payment in another, the job it was for in a third. Node & Nerve stores each record once and keeps how they relate, so a question can follow those links instead of guessing.
Connect what you use
Bring in records from the tools and files you already keep. Each one keeps a link to where it came from.
Records become a graph
Jobs, bills, payments, people, tasks and decisions become nodes, joined by what the records actually say. Nothing is inferred to fill a gap.
Ask, and get told
Ask in plain language or set what should notify you. Each answer lists the records it used and what is uncertain.
Integrations
What works today, and what doesn’t yet. A connection is only shown as working once records from it have reached the graph.
Ask questions and get answers, with evidence links, in a private chat.
Live, allow-listed chatsImport exports from accounting tools or spreadsheets. Every row is checked before anything is written.
Live, API and CLISend records, relationships and sources to the API. Re-sending the same record changes nothing.
Live, HTTP APIRead the sheets you choose after signing in with Google, through Composio.
In developmentThe account connects through Composio, but no read actions are available for it yet.
Waiting on connectorEvery finding says what kind of claim it is
Figures are computed by the engine from the graph, never by a language model. Each finding is labelled, so you can tell a recorded fact from a gap.
Recorded, and backed by a source.From the demo: CT-104 for Pon Logistics: 3 closed container bodies, ₹15,30,000, contracted 10 Sep 2026, expected by 31 Oct 2026; status in_production. Jobs: J-104A (painting), J-104B (fabrication), J-104C (fabrication).
Derived from facts by a stated rule.From the demo: Recorded direct cost so far: ₹8,51,700 from 27 book entries (35 charges, as some are split across bodies): materials ₹7,59,900, labour ₹71,500, machinery ₹11,800, other direct ₹8,500.
Not recorded. Never filled in as zero.From the demo: WG-416: Murugan, Welding, 4 days (2026-10-12 to 2026-10-15), ₹3,600 has no job recorded. It may belong to this or another contract; it is not counted anywhere.
Two records disagree.From the demo: PUR-2058 (Sri Balaji Steels, ₹58,000): the purchase book says 'paid', but the supplier payment book shows ₹0 cleared against it (payments: none).
A person has to decide before the figure is complete.From the demo: PUR-2053 (PU enamel paint 20L (3 drum), ₹28,500) is noted for CT-104, CT-105 with no split recorded. ₹28,500 is not counted in any contract's cost until the split is decided.
Answers in the chat you already use
The Telegram bot answers the owner’s questions from the same graph and links to the evidence. It only replies to chats the owner has allowed, and it never messages customers or suppliers.
- Commands answer straight from the graph, without AI.
- Questions in plain language go to an agent that can only call the investigations.
- Scheduled alerts for what matters to you come with workspaces.
Recorded direct cost (provisional): ₹8,51,700
Provisional profit ₹6,78,300 (44.3%)
- 4 items need your decision
- 5 items to verify
Try it on two fictional companies
The same engine, two very different businesses. Both run live against FalkorDB.
A truck-body builder’s contract costs
Purchases, wages, machinery and payments from separate paper books. What has contract CT-104 cost so far, and what is missing?
Open the contract investigationA software team’s late project
Tickets, incidents, owners and decisions. Why is Project Atlas late, who owns the blockers, and which decisions still apply?
Open the project investigation