Stop fixing spreadsheets.
Start shaping the plan.
Magpie is an AI-native finance workspace. Live data from every system, agents that build and update your models, and scenarios your whole team can actually trust.
No credit card · Import a spreadsheet and model in minutes
- Preparing customer details
- Analyzing historical financial data
- Forecasting revenue growth
- Assessing potential market risks
Profit Trend and Expense Impact
Created Revenue Forecast — Growth +30% · Q3. Increased new customers in EMEA by 30% while keeping churn unchanged, then recomputed active accounts and revenue under a dynamic ARPU model.
Ask something or @ mention a space
100+
prebuilt metrics
6
systems synced live
<50ms
full model recalculation
100%
of changes audited
Every number, and where it came from.
Variables, not cells. Each row carries its own formula, its own trend, and its own history — so nobody has to reverse-engineer a spreadsheet at quarter end.
Operating Profit Drivers
$1,230,569
$150,120
$423,112
Monthly Revenue Comparison
Profit Breakdown
Operating Profit Change
Typed variables
Currency, count, percent — each with its own aggregation rule, so quarterly and yearly rollups are correct by construction.
Formulas as structure
Formulas reference variables, not coordinates. Rename anything and nothing breaks.
Trends in place
A sparkline on every row means you see the shape of a number before you read the number.
One workspace, from revenue to runway.
Use Magpie Modelling across your entire planning workflow — the same engine, the same numbers, six different questions.
ARR Planning
Forecast recurring revenue, track churn and expansion, and model growth scenarios so GTM and finance stay aligned on targets.
Cash Flow Forecasting
Build real-time cash flow views that combine revenue timing, expenses, and collections to reveal your true liquidity picture.
Headcount Planning
Plan hiring by function, level, and location, layering in costs and timelines to see how headcount impacts runway and margins.
Capacity Planning
Integrate operational and financial data to understand capacity constraints, model utilization, and decide when to add resources.
Runway Forecasting
Project runway under different growth and spend scenarios so you know exactly how plans affect your next raise or profitability date.
Expense Management
Analyze expenses by team and vendor, flag overspend, and test cost-saving ideas before you commit to changes in the real world.
Built like an instrument, not a spreadsheet.
Multi-dimensional, no-code modelling
BuiltModel by product, region, channel, or any custom dimension. Dimensions belong to the variable, so a parent row is always a rollup of its members and can never disagree with them.
See every what-if in one place
BuiltScenario overlays that leave the base case untouched, and a grain switch from months to quarters to years that knows a balance takes the closing month while a flow sums. Side-by-side comparison and AI forecasting are next.
Let agents build with you
Partly builtIn reconciliation this is live: an agent adjudicates only what deterministic rules could not resolve, behind a gate that recomputes every number it returns. Agents that build models are designed and not yet built.
Bring your data in
Partly builtCSV ingestion across seven sources today, with every unparseable row recorded as a typed rejection rather than dropped. Live ERP, CRM, billing and HRIS connectors are on the roadmap.
Plan together, stay in control
Partly builtEvery change — human or agent — is a typed command carrying its own inverse, so undo and the audit trail are one mechanism. Comments, approvals and shared presence are on the roadmap.
Prebuilt metrics and reusable logic
RoadmapOne template exists today: a full ARR waterfall. The library of 100+ metrics and reusable components is the plan, not the product.
AI proposes. You decide.
Agents never write directly. Every change arrives as a reviewable proposal, and every number one returns is recomputed before you see it. That is live in reconciliation today — where the agent closes cases the rules cannot, without adding a single wrong match — and it is the same mechanism the modelling agent will use.
- Staged changesProposals render as ghost values next to the live ones.
- Compare before acceptDiff any proposal against the base case or another scenario.
- Reversible by designEvery accepted change is a command in the audit log — and every command has an inverse.
A model only sees what the rules could not settle.
A Next.js app on EC2 behind Caddy, Postgres on RDS, and a reconciliation pipeline built so that every step a model takes is followed by plain code that checks it. Here is how the pieces fit, and the AWS services around them.
01 Serving path
Browser
magpie.akkki.tech DNS on Vercel
Caddy
HTTPS 443, cert auto-renewed by Let's Encrypt
Next.js 16 app
server components, route handlers, Better Auth
RDS Postgres
private subnet, no public IP Prisma 7, SSL
02 Reconciliation, inside the app
Statement files
7 CSV sources: bank, settlements, ledger and more
Ingest
pure function of file contents. Bad rows kept.
Rule matcher
exact, tolerance, structural tiers; explainable
Adjudicator
LLM API call, structured output, ranked candidates
Validation gate
recomputes every number; rejects ungrounded ids
Review queue
a person confirms proposals and exceptions
Rules-only run on the synthetic batch: 11,269 records, 419 of 456 results auto-applied, 8.1% escalated. 100% precision, 0% false matches.
03 Finance-ops agent, inside the app
A question
asked in plain English
Supervisor
LangGraph deep agent, plans as a todo list, holds no read tools
Analyst subagents
model-analyst and data-analyst: read-only, run in parallel
Approval gate
the graph halts before any write tool runs
Proposals
staged changes a person accepts or rejects
04 Around it, on AWS
ECR
container image, pulled at boot
Secrets Manager
runtime env, read by the instance role
S3
statement archive; private, versioned, encrypted
CloudWatch
container logs, 6 alarms, dashboard
SNS
alarm emails
The instance role reaches one bucket and these log groups, nothing else. Secrets never appear in the launch config.
Your model is only as fresh as its inputs.
Connect the systems that already hold the truth. Magpie syncs continuously and writes every change to the audit log, so a number moving overnight is always explainable.
Close the books on spreadsheet planning.
Bring your data, your drivers, and your team into one workspace — and let the agents handle the setup.