The open-source knowledge graph for finance
Finance AIthat ties to the books.
RoboSystems is an open-source knowledge graph that gives your AI agents your general ledger and public companies' filings as structured, traceable data, on your infrastructure, with the model you choose.
Built by an inactive CPA and ex-Workiva knowledge engineer. Our own companies' books are kept on it.
- Open source (Apache 2.0 and MIT)
- Every number traced to its entry
- Your cloud, your server, or ours
- Claude, ChatGPT or any MCP client
- SOC 2 in progress
What it's for
The numbers tie. The books stay put. The filings sit beside them.
Every number opens to the entries behind it.
The numbers don't tie
Your finance agent needs the books, not a CSV export: the chart, the calendar, the close rules, every number traced to its entry.
Book 20 minutes →The books can't leave
AI on your clients' books without the books leaving your control. Client confidentiality, data residency, or a legal team that won't send the ledger to a vendor's AI. Run it in your cloud or on your own server, with the model you choose. Your data trains no one's model.
Book 20 minutes →The filings and the books live apart
Your ledger and 8,000+ public companies' XBRL filings, in one ontology, with every number traced back to the entry or the tagged fact.
Book 20 minutes →Three kinds of graph, one platform
One ontology underneath: the RoboSystems graph model, labelled in XBRL's standard terms wherever XBRL has one. Your AI reaches all of it over MCP, and every answer opens to its entries or its tagged facts.
Your ledger
RoboLedger's books mapped onto the ontology, with the fiscal calendar and the close rules.
- •Chart of accounts mapped to XBRL concepts
- •Period close, schedules and reconciliations
- •Every entity of the group in one graph
- •Every number traced to its entry
The public record
8,000+ SEC filers' XBRL filings in the same schema as your ledger, with search across MD&A, risk factors and notes.
- •Full-text and semantic search with hybrid scoring
- •Search MD&A, risk factors, and iXBRL disclosures
- •Statements rendered from the XBRL, not PDF scans
Your own graphs
Bring a schema, load Parquet, query with Cypher. LadybugDB underneath.
- •Your own node and relationship types
- •DuckDB staging for data validation
- •Columnar storage for fast aggregations
- •LanceDB vector indexes for semantic search
Subgraphs and memory
A subgraph is a workspace beside your graph where your AI adds node and relationship types and writes Cypher over MCP. Semantic memory carries its work between sessions.
- •Works with Claude, ChatGPT or any MCP client
- •Schema introspection and typed GraphQL reads
- •OAuth sign-in or API key authentication
API & SDKs
The platform REST API: GraphQL for reads, named operations for writes.
- •Extensions GraphQL with per-graph schema introspection
- •Named operations with idempotency and audit trail
- •TypeScript and Python SDKs with full types
- •SSE streaming for real-time operation progress
Security & Isolation
A dedicated graph-database instance per subscription, or a Dedicated Deployment: a segregated AWS account RoboSystems runs, transferable to you.
- •Tested cross-tenant isolation
- •Encryption at rest and in transit
- •Passkeys, plus SSO and SCIM on Dedicated Deployments
- •SOC 2 Type II compliance in progress · Deployment options · Trust Center
Run it your way
Let us run it, or run it in a segregated AWS account we operate under our SOC 2 program and hand over when you say. Both run frontier models. Or self-host on a box you own, on bare metal, or in your own AWS, where an open-weight model is an option, with help from Harbinger FinLab if you want it. Open source (Apache 2.0 and MIT): you keep the code and the data whatever happens to any vendor, including us. Nothing held hostage.
RoboSystems GitHub
Three kinds of graph on one platform: your ledger, the public filings, and graphs you model yourself, from your own schema, loaded from Parquet and queried with Cypher. The main repository holds all of it: the API server, the infrastructure templates and the deployment workflows, to run locally or in your own cloud.
Why Open Source?
- It's Yours
Fork the repository and you own the software: nobody can reprice it, take it away, or shut it down. Stay close to the main line and every fix, integration and reporting update comes with it. Why we build it this way
- Semantic Sovereignty
Own what your numbers mean. Definitions, taxonomies, and calculation logic are open, inspectable artifacts, because a black box in your own cloud is still a black box. Read the essay
- Security Through Scrutiny
Open code means more eyes finding and fixing potential issues.
- One ontology underneath
The RoboSystems graph model, labelled in XBRL's standard terms wherever XBRL has one. Every graph runs on it in LadybugDB, and reports leave as a holon (RDF), XBRL 2.1 or Tavi.
Three ways to run it
Point its AI at Claude, ChatGPT or any MCP client, or, when you host it yourself, an open-weight model on your own hardware.
- •Managed by us: from $99 a graph. Pricing
- •Dedicated: a segregated AWS account, quoted per engagement. Deployment options
- •Self-host: free. See Self-Hosting on a Server and the Bootstrap Guide.
- •Self-host help: Harbinger FinLab sets it up with you, on a screenshare you control.
Fork & Deploy
Deploy your own instance with secure GitHub OIDC, with no AWS credentials stored in GitHub. Full CloudFormation templates and GitHub Actions workflows included.
Bootstrap GuideDownload RoboSystems
Client SDKs
Official clients for the RoboSystems API.
Local Development
# Clone and install tools
git clone https://github.com/RoboFinSystems/robosystems
cd robosystems
brew install uv just
# Start local Docker stack
just start
# API at http://localhost:8000For production deployment, see the Bootstrap Guide.
The Finance AI Deployment
RoboSystems where you choose (the RoboSystems managed platform, a Dedicated Deployment RoboSystems runs, or your own cloud), on your books, with three working agent workflows, in two to three weeks, at a fixed fee.
- Variance explanation, close checklist and peer benchmark in your MCP client, every number traced
- Your general ledger and the public companies you compare yourselves against, read side by side. QuickBooks connects directly; any other ledger comes in through the API or a file.
- Your team runs the deploy with Harbinger on a screenshare you control; workflows are built through keys you issue and can revoke
- Your people approve every entry. Harbinger never signs in to your books.
- $7,500 to $10,000, fixed by scope: half up front, half at handover.
- Hosting is separate: from $99 a month on the managed platform, Dedicated quoted, or free to self-host. Peer benchmarks use the SEC repository ($29 or $99 a month, 14-day trial).
- You keep everything Harbinger builds.
Thursday or Friday afternoon not good? and we'll find a time.
Integrators and consultancies: deliver the Finance AI Deployment for your clients. You keep the services; RoboSystems runs the platform.
What's built on it
Products that run on the same platform and the same public API.
How It Works
API
Graph
RoboLedger keeps the books in the ledger graph. RoboInvestor tracks positions beside the SEC repository. Both expose MCP tools.
Inside your company
RoboLedger
The general ledger built on RoboSystems
A full accounting system in its own right: close, consolidation, schedules, bank feeds, reconciliations and forecasting, every number traced to the entry.
Connections
- •QuickBooks over OAuth2 with on-demand sync
- •QuickBooks MCP server for Claude, ChatGPT, or any MCP client
- •Chart of accounts, journal entries, and transactions
- •Trial balance with full debit/credit detail
Ledger, Close & Reporting
- •Event-driven inbox: AI pre-classifies, you approve
- •AI auto-maps chart of accounts to US-GAAP with confidence scores
- •Live Balance Sheet, Income Statement, and Cash Flow: no close required
- •Guided period close: fiscal calendar, depreciation and prepaid schedules, rule-engine gating
- •Report Creator: XBRL 2.1 export with publish lists
Planning & Analysis
- •Plan: actuals and forecast in one monthly grid
- •Block Explorer: any number down to the facts and rules behind it
Agents & MCP
- •AI Console: ask about your books, or have it draft and sort the routine work for you to review
- •Your AI assistant drives the close over MCP: author scenarios, define metrics, write handler rules
- •Push approved entries back to the source system
Outside your company
RoboInvestor
Investment Portfolio & AI Research
Equity research on public companies, your fund's positions with their terms, and the reports portfolio companies share from RoboLedger. Building next: positions linked to the ledger, brokerage sync and portfolio analytics.
Track public and private positions in one book, research any public company against the SEC knowledge graph, and read equity research generated from the filings.
Portfolio Management
- •PE & VC positions with full terms: SAFEs, notes, LLC units, LP interests, warrants, options, and RSUs
- •Cost basis, valuations, and holdings rolled up per issuer
- •Cross-graph linking: a holding connects to its issuer's own graph
SEC Research Console
- •75,000+ XBRL filings across 8,000+ public companies
- •AI-powered deep research: plain-English questions become graph queries
- •Statements rendered from the XBRL, not PDF scans
- •Full-text and semantic search across filings and your own documents
- •MCP tools for Claude, ChatGPT or any MCP client
Equity Research
- •One company per report, every figure traceable to its filing
- •Narrated video brief alongside the written report
xbrlkit: a free reader for XBRL reports, local, with its own MCP server.
SEC XBRL Repository
The public record, 14 days to try
8,000+ public companies' XBRL filings in a knowledge graph, in the same schema as your ledger, with full-text and semantic search across MD&A, risk factors and notes. Connect from Claude, ChatGPT or any MCP client, with nothing to build.
MCP Clients
Use with Claude, ChatGPT or any MCP client, no install required.
MCP Protocol
Paste one URL and sign in.
https://api.robosystems.ai/v1/graphs/sec/mcpAPI & SDKs
npm i @robosystems/clientpip install robosystems-clientBuild on RoboSystems
Connect your data sources, write your own integrations, and build applications and agents on the same public API that RoboLedger and RoboInvestor run on.
Official Integrations
QuickBooks
Accounting
Sync your chart of accounts, transactions, and trial balance. Full OAuth integration with on-demand sync.
Need a QuickBooks account? →SEC XBRL
Public Filings
XBRL facts from 8,000+ public companies in a shared graph, full-text and semantic search across the filings, and the Analyst Operator for plain-English questions. Updated daily, nothing to import.
See the SEC Repository →Mercury
Business Banking
Connect your Mercury accounts with OAuth. Posted transactions land in your ledger's inbox with a suggested category, ready to review and post.
Plaid
Bank Feeds
Link bank and card accounts at other institutions. Transactions flow into the same inbox, and nothing posts to the ledger until you approve it.
Build Your Own Integrations
Connect any source without touching the platform. An integration is a small program in its own repository that writes to RoboSystems through the public API with an API key. It never runs inside the platform, so it survives every release, runs identically against the managed cloud or a self-hosted deployment, and your source credentials stay on your side.
Ledger
Business events from any system you run (invoices, payments, payroll) landed in an inbox for approval.
What we enforce
- •Double-entry balance
- •Capture, then approve
- •Closed-period gate
- •Idempotent per source event
Semantic facts
A custom vocabulary, then observed values per period: usage counts, operational KPIs.
What we enforce
- •Typed concepts
- •Presentation structure
- •Replace per period
- •Provenance on every fact
Raw graph
Parquet or CSV files, staged and materialized into graph nodes and relationships.
What we enforce
- •Per-graph schema
- •Bulk ingestion pipeline
- •Staging you can query
Nothing to stand up: the template ships a GitHub Actions schedule and a Claude Code setup: fill in collect and transform, set two repo secrets, and it runs.
Start from the integration template, see a working example, or read the integration guide.
Clients & SDKs
The same surface the applications above are built on. Every graph is a remote MCP server, and typed clients cover the REST API end to end.
MCP Endpoint
Every graph is a remote MCP server: one URL, sign in, no install
https://api.robosystems.ai/v1/mcpTypeScript Client
Full-featured client for JavaScript and TypeScript applications
npm install @robosystems/clientBuilding an application on RoboSystems?
Talk it through in 20 minutes
Tell us which job you have. We'll show the trace on a demo company and say plainly whether the Finance AI Deployment fits.
Thursday or Friday afternoon not good? and we'll find a time.
Already have an account? Login