
Digits charges $65 a month for its cheapest plan and says its ledger books more than 95% of transactions on its own. Rillet's buyers on Vendr pay a median of about $28,250 a year, plus a $10,000 implementation fee. Both get called "AI bookkeeping software," and I think that label hides the only decision that matters: which ledger do you trust to hold the books when an agent is doing most of the typing?
I've gone through the five AI-native ledgers controllers ask me about most, Digits, Puzzle, Campfire, Rillet and Light, and held them against the two ledgers they're trying to replace, QuickBooks Online and Xero. Prices and agent access come from the vendors' own pages where they publish them. Where they don't, I say so.
| Ledger | Who it fits | Price | Multi-entity | Agent access (MCP) |
|---|---|---|---|---|
| Digits | US small businesses, startups, CAS firms | $65 / $100 / $250 a month per business | Multi-entity since Oct 2026; consolidation through Reach Reporting | Read and write (38 tools) |
| Puzzle | Early-stage US startups | $24 / $58 / $96 / $288 a month, billed annually | Consolidation through Joiin; USD only | Read-only by default, writes opt-in |
| Campfire | Venture-backed SaaS leaving QBO or NetSuite | Quote only | Native | 12 read-only tools; writes through the REST API |
| Rillet | Multi-entity B2B SaaS | Quote only (Vendr median about $28,250/yr) + $10k implementation | Native, with eliminations | Read and write, scoped to the user's permissions |
| Light | International groups needing local and group books | No published price | Native, multi-book | Hosted MCP exists; tool list not published |
| QuickBooks Online | The default US small-business ledger | Free, or $38 / $85 / $140 / $340 a month | One subscription per company | Open-source MCP in early preview (about 144 tools) |
| Xero | Small businesses that want no per-user fees | $27 / $59 / $97 a month (list) | One organization per entity | Open-source MCP, 51 tools (26 read, 25 write) |
An AI-native ledger keeps transactions, contracts and vendor records in one schema it controls, so an agent can trace a GL line back to the Stripe charge or the invoice PDF without a sync in between. An add-on reads your QuickBooks or Xero data out over an API, works on it and writes it back.
That sync is where I've watched books drift. A categorization tool and a receipt app and a rev rec spreadsheet all touching the same ledger means three places a mapping can go stale. Still, I'd push back on anyone who tells a single-entity US company with a happy outside CPA to rip out QuickBooks. If your tax preparer lives in QBO and your revenue is simple, the migration risk can easily outweigh the cleaner data model. I'd compare the two baselines first on our QuickBooks Online vs Xero page before shopping for a replacement.
Puzzle, at $24 a month billed annually, and Digits, at $65 a month with no seat fees, are the two self-serve options with prices on the page. Everything above them is a sales call.
Digits launched its Autonomous General Ledger in March 2025 and publishes three tiers: $65, $100 and $250 a month per business (Digits pricing). It serves US-based entities only. The 95%+ auto-booking figure is Digits' own claim, so I'd test it on a month of your real bank feed before believing it. Multi-entity arrived in October 2026, but consolidated reporting still runs through Reach Reporting.
Puzzle runs $24, $58, $96 and $288 a month billed annually, and its AI Close feature ships on every plan. It's USD only, and consolidation goes through the Joiin add-on. If you're a pre-seed founder with one entity and one currency, Puzzle is the cheapest real ledger here. If you've got a Canadian sub, it's already the wrong tool.
You need one of them when intercompany eliminations or ASC 606 contract modifications are living in a spreadsheet next to your ledger. That's the moment the $65 tools stop being cheaper.
Campfire is quote-only. Its revenue automation handles subscriptions, usage, standalone selling price allocation and contract modifications natively. Campfire says its Migration Agent moves GL history "in minutes," and our review puts that at about 15 minutes. I'd budget far longer than that for tying out the result.
Rillet is also quote-only. Vendr data puts the median contract at about $28,250 a year, with a range of roughly $20,100 to $34,800, and Rillet charges a $10,000 implementation fee. TechCrunch reported 600+ customers in August 2026. It does native multi-entity with automated eliminations and its own ASC 606 engine.
Light is built in Copenhagen for international groups, and its multi-book setup lets one transaction feed a local statutory book and the group book. Light doesn't publish pricing. One third-party estimate puts it somewhere between $35,000 and $150,000 including implementation, which is too wide to plan around, so get a written quote.
It ranges from read-only to full write access, and this changed fast during 2026. Campfire's MCP server exposes 12 read-only tools and keeps journal writes on its REST API. Puzzle's MCP is read-only until you opt in to writes. Rillet's lets an agent read and write within the connected user's permissions.
Digits is the one to watch. It launched its MCP server read-only in April 2026, and its current help doc lists 38 tools, 19 of them write tools. It also has a tool to propose a change to a locked period instead of posting straight into one, which I like. On the baselines, Intuit's open-source QuickBooks MCP server is labeled early preview with roughly 144 tools, and Xero's has 51, split 26 read and 25 write.
My take: write access is a feature only if your permissions are tight. I'd connect any of these with a read-only user first, run a full close with the agent proposing and a person posting, and only then hand it a write-capable token. We go deeper on that test in asking whether an agent can actually run your software.
Run one full month in parallel and don't cut over until the new trial balance ties to the old one to the penny. Everything else on the checklist supports that one step.
If you'd rather hand the books to people than to software, read Pilot vs Bench first. And if you're the accountant deciding which of these tools is worth learning, our software demand data shows which ones employers actually name in job postings.
For a single US entity in dollars, Digits ($65 to $250 a month) or Puzzle ($24 to $288 a month, billed annually). For multi-entity SaaS with real ASC 606 work, Rillet or Campfire, and Light if you run statutory books in several countries.
Not yet. Digits' own figure is 95%+ of transactions auto-booked, which still leaves exceptions, accrual estimates and standalone selling price judgments for a person to review and sign off.
Rillet, Digits, QuickBooks Online and Xero all have MCP servers with write tools today, and Puzzle allows writes once you opt in. Campfire keeps its MCP read-only and routes journal writes through its API.
Puzzle is USD only and Digits serves US-based entities only. Campfire, Rillet and Light are the ones built for multi-entity groups, and Light's multi-book design targets companies with entities in several countries.
If you're seriously weighing a switch, try this: export last month's GL, give an agent read-only access to it and one of these ledgers' sandboxes, and ask it to rebuild your close checklist. You'll learn more about the tool in one afternoon than from any demo.