Light Review
Light is a Copenhagen built, AI native ERP that runs multi entity, multi currency finance on one live ledger, deriving each entity's local statutory book and the group book from the same transactions across 21 countries, and it is aimed at multi entity scale ups and mid market groups replacing a legacy ERP or a stack of single entity accounting tools.
Independently scored, one dimension at a time. Every dimension is sourced, and we say whether a claim is published, reported, or estimated.
The API documentation gives an MCP endpoint at api.light.inc/rest/ext/mcp that takes a personal lmcp token, and describes roughly a hundred tools covering search and read across master data, documents and reports, plus create, submit, approve and flagged write operations. Claude Desktop and Claude.ai connect through the Light connector with a Light account and no token. The Flash number 10 post of 13 August 2026 claims Light passed Anthropic's review to become a built in Claude connector and that Custom Agents shipped to every company, both vendor claimed. The deduction is that MCP acts as the signed in person rather than as a service account, so every autonomy claim still rests on the roles a human already holds.
Multibook derives statutory and group books from the same transactions, and the consolidation page describes intercompany lines flagged at posting and reversed on a dedicated elimination ledger, with three currency amounts stored on every ledger line and group translation posted to CTA under IAS 21 and ASC 830. Revenue recognition under IFRS 15 and ASC 606 ships inside the subscription management module. SOC 1 Type II was examined by AAFCPA for 1 July to 31 December 2025 and SOC 2 Type II by Prescient Assurance for 10 February to 1 June 2025, both unqualified with no exceptions noted. The deduction is that accounting periods are generated on demand rather than automatically, so the API documentation states that a date no generated period covers is unprotected and postings into it are allowed.
The consolidation page describes the consolidated profit and loss and balance sheet as live views of the ledger rather than a month end step, with an eliminations column computed from postings that drill back to real entries. The close runs as per entity tasks, lock AP, lock AR and lock JE as soft locks, then FX revaluation, then closing the period in chronological order, and year end posts a year closing entry per entity. Reporting is offered as pivot tables, direct SQL and charts in chat, plus Google Sheets, Slack and Teams surfaces. Tillo reports a 12 day close falling to 5 and Alva Labs a same day close target.
The API index describes 193 endpoints across 31 resources on api.light.inc, and the published OpenAPI specification carries 143 paths and 191 operations. Rate limits are documented at 300 requests a minute per user and 100,000 a day per organisation, with the 429 response headers spelled out. Authentication is an API key or an OAuth 2.0 authorization code flow, and the documentation states that PKCE is not supported. The deduction is webhooks: the word does not appear once in the documentation index or in the OpenAPI specification, so an integration has to poll.
The Tillo customer story puts 4 trading and group entities, 23 countries, 23 currencies and 5 to 15 million monthly transactions on one Light ledger. Bill pay is described as covering 80 plus countries, and 21 country pages carry local statutory books, tax and e-invoicing. The ceiling is the segment rather than the architecture: ERP Research, reviewed 3 August 2026, puts the ideal fit at 51 to 5,000 employees and records no inventory, manufacturing or warehouse module and no on premise option.
Light's own post on ERP implementation timelines dates Dreamdata live on all 3 entities on 1 April 2026, 6 weeks from cutover, then tells buyers at that size to plan for a few months rather than a few weeks. ERP Research lists a typical go live of 2 to 12 weeks and a total project cost of 35,000 to 150,000 US dollars including implementation. All Gravy moved six years of history back to January 2020, roughly 200,000 transactions in the Denmark entity alone. Native import paths are published for QuickBooks, Xero, e-conomic, Fortnox, Visma and Billy.
The site indexes a 173 article help centre, and status.light.inc lists five components with RSS and a JSON API. The API documentation routes higher rate limit requests to help@light.inc, and ERP Research records implementation as delivered by Light's own team. No response time target, support tier or service level agreement appears anywhere on the public site, and the EuropeanStack review of 25 July 2026 counted only two G2 reviews, so independent evidence of day to day support quality is close to absent.
There is no pricing page: light.inc/pricing returned 404 when fetched on 6 September 2026, and no pricing URL appears in the sitemap. ERP Research and EuropeanStack both record that Light publishes no rate card by user, entity or module. The Terms of Service place every fee in an Order Confirmation, renew each Service Period automatically for a successive period of equal length unless three months written notice is given, and state that payments already made are not refundable.
Written by the Audit Friendly research team. No vendor edits, no sponsored placement.
What Light sells is one live ledger for a group that has more than one legal entity. Multibook derives each entity's statutory book and the group book from the same transactions, intercompany lines are flagged when they post and reverse onto a dedicated elimination ledger, and every ledger line stores the transaction currency, the entity's functional currency and the group currency fixed at the posting date rate. Twenty one country pages carry the local reporting standard, the tax return and the national e-invoicing network, and revenue recognition under IFRS 15 and ASC 606 sits in the same platform as accounts payable, accounts receivable, corporate cards and spend. The agent surface is real rather than a chat box: the API documentation publishes an MCP endpoint offering roughly a hundred tools that read, create, submit, approve and write.
The catch is that almost nothing commercial is public. There is no pricing page: light.inc/pricing returned 404 on 6 September 2026, and every figure a buyer sees comes from a quote. The company was founded in 2022, crossed 100 full time employees in mid 2026 and has raised 43 million US dollars, which ERP Research describes as the smallest funding base among the AI native vendors it tracks. Independent review evidence is close to absent, and the public REST API documents no webhooks at all, so an integration polls inside a ceiling of 300 requests a minute per user. Light is also financials only: ERP Research records no inventory, manufacturing or warehouse module and no on premise option.
Four things belong in writing. Ask for the current SOC 1 Type II and SOC 2 Type II reports and read the periods, because the published SOC 2 window ends 1 June 2025 and covers the Security criteria alone. Ask how accounting periods will be generated for every entity, because the documentation states that a date no generated period covers is unprotected and accepts postings. Negotiate the renewal, because the standard Terms of Service renew each Service Period automatically for a successive period of equal length unless three months written notice is given, payments already made are not refundable, and account information is deleted 90 days after the agreement expires. And settle the governing law: the contracting party is Light Company ApS in Copenhagen, registration number 43523503, with Danish law and the Copenhagen courts holding exclusive jurisdiction.
- A group of two to twenty legal entities that needs local statutory books and one consolidated close, with intercompany eliminations and FX handled in the ledger rather than in a spreadsheet.
- A finance team that wants an AI agent to read and write against the books through a documented MCP surface of roughly a hundred tools, with a personal token per user and every action inside the caller's own roles.
- A European or globally distributed company that needs local tax and e-invoicing rails such as XRechnung, Italy's SdI, Peppol and MTD VAT in the same system as the group ledger.
- A buyer who needs published, comparable pricing during evaluation, since Light has no pricing page at all and quotes every deal through a demo.
- A manufacturer or distributor that needs inventory, warehousing or production, which ERP Research records as absent from Light's documented capability set.
- A controller who wants a long operating history and a deep independent review base behind the general ledger, since Light was founded in 2022 and its public review footprint is tiny.
The jobs a controller actually hires Light for
Jobs, mapped to how finance teams actually buy.
Watch the same task in each tool
Forget the 40-row feature grid. Pick a finance workflow and see how this tool and its closest peers actually do it, with our verdict on who wins for whom.
Compare on what matters
What it actually costs
Here's what it really costs: published rates where a vendor lists them, aggregated quote data where they don't, plus a path to your own number.
Light publishes no pricing tiers of any kind. light.inc/pricing returned 404 on 6 September 2026 and no pricing URL appears in the sitemap. The Terms of Service put every fee in an Order Confirmation agreed with sales.
The analyst site ERP Research, reviewed 3 August 2026, lists a starting price of about 35,000 US dollars a year and a total project cost of 35,000 to 150,000 US dollars covering software plus implementation. This is an estimate by a lead generation site, not a figure Light publishes.
The ledger, Multibook statutory books, consolidation, accounts payable, accounts receivable, subscription management with IFRS 15 and ASC 606, spend management, corporate cards, global bill pay across 80 plus countries, reporting and the Slack, Teams, Google Sheets and mobile surfaces are all presented as one product rather than separately priced modules.
The Terms of Service renew each Service Period automatically for a successive period of equal length unless three months written notice is given to expire at the end of the current period. Payments already made are not refundable except on termination for cause, and Light deletes all account information 90 days after the agreement expires.
What it actually takes to go live
The API documentation on periods and locks states that accounting periods are generated on demand by the company rather than automatically, and that a date no generated period covers is unprotected, so postings into it are allowed. It advises treating an unexpected success on a far future or far past date as something to raise. Agree in the implementation plan who generates periods ahead for every entity, and confirm it during the first close rather than assuming the lock is always on.
The word webhook does not appear once in the documentation index at light.inc/docs/llms.txt or in the published OpenAPI specification, which carries 143 paths and 191 operations. Any downstream system that needs to know a bill was approved or a period closed has to poll, inside a limit of 300 requests a minute per user and 100,000 a day per organisation that resets at midnight UTC. Confirm the event delivery story in writing before scoping a real time integration.
Clause 13 of the Terms of Service renews each Service Period automatically for a successive period of equal length unless terminated with three months written notice to expire at the end of the current period. Clause 8.1 states that payments already made are not refundable except on termination for cause. Clause 13.4 states that Light deletes all account information 90 days after the agreement expires, so the data extraction has to be planned inside that window. The contracting party is Light Company ApS in Copenhagen, registration number 43523503, and clause 20 puts the agreement under Danish law with the Copenhagen courts holding exclusive jurisdiction.
light.inc/soc-2 records a SOC 2 Type II examination by Prescient Assurance covering 10 February to 1 June 2025, Security trust services category only, unqualified with no exceptions. light.inc/soc-1 records a SOC 1 Type II examination by AAFCPA covering 1 July to 31 December 2025 on the AI Accounting Platform system, also unqualified. Both reports are confidential and released under a non disclosure agreement. Ask for the current period on both before an auditor relies on them, and check the complementary user entity controls the SOC 1 report names.
ERP Research, which reviewed Light on 3 August 2026, records that Light is a financials only platform with no manufacturing and no warehouse management module, and that inventory and supply chain do not appear in its documented capability set at all. It also records that Light is cloud only on EU hosted AWS with no on premise or self hosted option, and puts the ideal fit at 51 to 5,000 employees. A group that holds physical stock as a core operation needs a second system for it.
Scored on what an outside agent can actually reach: the documented API, how it authenticates, and whether the vendor ships an MCP server.
Light publishes its agent surface in the same documentation as its REST API, which is unusual in this category. The MCP server sits at api.light.inc/rest/ext/mcp and speaks JSON-RPC 2.0 over HTTP POST, with initialize, tools/list, tools/call and ping. Credentials are separate from the REST API: a personal token with an lmcp prefix, created under Settings, Profile, MCP Tokens, shown once, up to 20 active per user with an optional expiry. Claude.ai and Claude Desktop connect through the Light connector with a Light account instead of a token. API keys cannot call the MCP server and MCP tokens work only on the MCP endpoint.
The documented surface is roughly a hundred tools: search and read across master data, documents and reports, create and submit, approve, and a set of flagged write operations. The tool list is filtered at runtime to what the caller may do, and amounts come back in major units with field names ending in InMajors rather than the minor units the REST API uses. The server also exposes Light's help articles as tools, so a how to question about the product is answerable in the same session.
Around it Light publishes Custom Agents, described as a recurring finance job briefed in plain language that then runs on a schedule under its own identity inside company policies and the audit trail, and two free MIT licensed CFO agent packs of 11 agents and 4 workflows for Claude Code. Light states in its Flash number 10 post of 13 August 2026 that it passed Anthropic's review to become a built in Claude connector, and in its July 2026 highlights that its AI handled more than 50,000 admin jobs in July with 96 percent running unprompted and 99.4 percent of customer ledger entries automated. Every one of those figures is vendor claimed and none is independently reported.
Close the month
Who wins for whom: Several entities you want to see consolidated live: Light. A close enforced by a proven checklist at scale: NetSuite. Variance commentary written for you, if you can do without a documented statutory reporting pack: Campfire.
Light
Consolidation as a live view. Consolidated statements are live views of the ledger with a computed eliminations column, and the close runs as per-entity tasks with soft locks.
Where it loses: Financials only, with no inventory or manufacturing, and the contract renews itself unless you give three months notice.
- Reporting & close: 80/100
- Time to live: 6 weeks to months
Source: AF score rationale: Reporting & close 80; Light watch-outs.
Watch: Alva Labs: From a Five-Day Close to a Same-Day Close 2:18 · Customer story · Light Streaming on YouTube (official)
NetSuite
A checklist that enforces the close. The Period Close Checklist sequences and enforces close tasks, and saved searches cover most controller reporting without leaving the system.
Where it loses: Reviewers report slowness at month end as custom scripts and searches pile up, and implementation runs from three months to well past a year.
- Reporting & close: 80/100
- Time to live: About 100 days
Source: AF score rationales: Reporting & close 80, Implementation 50.
Watch: Simplify Month-End Close in NetSuite 1:51 · Consultant walkthrough · Third-party consultant walkthrough
Campfire
A checklist plus a flux agent. A built-in close checklist, reporting that drills to the transaction, and a flux analysis agent that writes variance commentary.
Where it loses: No consolidated statutory or local GAAP reporting pack is documented, and it publishes three different go-live timelines at once.
- Reporting & close: 65/100
- Time to live: 4 to 12 weeks
Source: AF score rationales: Reporting & close 65, Implementation 71; Campfire watch-outs.
Watch: From 20-Day Close to 6: How Boulder Care Runs Finance with Campfire 2:27 · Customer story · Campfire AI on YouTube (official)
Reconcile the bank
Who wins for whom: European and US banks on one ledger, with FX-aware matching: Light. Reconciliation as the month runs: Campfire. Matching rules you can extend, with daily automated imports: NetSuite.
Light
Four feeds, two rule sets. Feeds from GoCardless in Europe, Plaid in the US, Stripe and host-to-host AMC, with system and user matching rules, duplicate feed lines excluded and FX-aware matching.
Where it loses: System rules can be switched on or off but not edited, so anything unusual needs rules of your own, and there are no webhooks for downstream automation.
- API & automation: 76/100
- Time to live: 6 weeks to months
Source: Light help centre, bank integrations and reconciliation automation rules; AF score rationale: API & automation 76.
No public demo of this task. Ask for a sandbox run into your own system before you sign.
NetSuite
Rule-driven matching at scale. Match Bank Data applies three default matching rules plus any you write, and automated imports can run daily through a bank connectivity SuiteApp.
Where it loses: Each import tops out at 10,000 transactions and 500 accounts, and bank connectivity is a SuiteApp or partner setup rather than on by default.
- API & automation: 78/100
- Time to live: About 100 days
Source: Oracle NetSuite help on bank data import and reconciliation rules; AF score rationale: API & automation 78.
Watch: NetSuite bank matching and reconciliation: system rules 6:21 · Consultant walkthrough · Fusion CPA on YouTube
Campfire
Reconciles through the month. Bank and card accounts reconcile as the month runs rather than at month end, with AI categorisation that learns from your data and approvals by dollar threshold.
Where it loses: The 90 percent time-saved figure has no published method, and the documentation does not spell out how banks connect.
- API & automation: 70/100
- Time to live: 4 to 12 weeks
Source: Campfire product demo (2026) and llms.campfire.ai; AF score rationale: API & automation 70.
Watch: Campfire product demo (2026), transactions and bank reconciliation 9:16 · Vendor product demo · Campfire AI (official)
Light, kept current
Light FAQ
Light publishes no price. The pricing page does not exist: light.inc/pricing returned 404 when fetched on 6 September 2026, and no pricing URL appears in the sitemap. Every plan is quoted, and the Terms of Service place fees in an Order Confirmation rather than a public rate card. The only public figure comes from the analyst site ERP Research, which reviewed Light on 3 August 2026 and lists a starting price of about 35,000 US dollars a year and a total project cost of 35,000 to 150,000 US dollars including implementation. Treat that as a third party estimate rather than a vendor published number, and get a scoped quote against entity count and country footprint before budgeting.
The standard Terms of Service run for the Service Period set out in the Order Confirmation and renew automatically for a successive period of equal length. Terminating requires three months written notice to expire at the end of the current Service Period unless something else is expressly agreed. Payments already made are not refundable except where the customer terminates for cause, in which case a pro rata refund applies for the remainder of the term. Light states that it deletes all account information 90 days after the agreement expires, so plan the data extraction well inside that window. The contracting party is Light Company ApS in Copenhagen, registration number 43523503, and the agreement is governed by Danish law with the courts of Copenhagen holding exclusive jurisdiction.
Light's own team runs the migration rather than a partner network. Light's post on ERP implementation timelines dates Dreamdata live on all three entities on 1 April 2026, six weeks from cutover, and then tells buyers at that size to plan for a few months rather than a few weeks. ERP Research lists a typical go live of two to twelve weeks. The variable is how much history moves and how much structure gets rebuilt: All Gravy moved six years of transactions back to January 2020, roughly 200,000 in the Denmark entity alone, and rebuilt its chart of accounts on the way. Native import paths are published for QuickBooks, Xero, e-conomic, Fortnox, Visma and Billy.
Yes, and further than most ERP systems allow. Light publishes an MCP server at api.light.inc/rest/ext/mcp that takes a personal token with an lmcp prefix, up to 20 active tokens per user, and Claude.ai and Claude Desktop can connect through the Light connector with a Light account instead of a token. The documentation describes roughly a hundred tools covering search and read across master data, documents and reports, plus create, submit, approve and flagged write operations, with the tool list filtered at runtime to what the caller may do. Two limits matter. The agent acts as the signed in person and inherits that person's roles, so it can never exceed them. And credentials do not cross over: API keys cannot call the MCP server and MCP tokens work only on the MCP endpoint. Light's statement that it passed Anthropic's review to become a built in Claude connector is vendor claimed, from its Flash number 10 post of 13 August 2026.
Light alternatives
Ask us anything
Answered only from our own published research on this tool, never from general internet noise. If we cannot answer it well, our research agents will dig in and publish a sourced answer.