The product
One platform for the .xlsx lifecycle — upload to allocate.
Ten things Eucalyptus does, in the order you'll use them. No webinars, no lock-in — the runtime, the dashboard, regression tests, usage allocation and export, self-host, enterprise identity, and an MCP server your agents can drive.
Interface samples use illustrative workspace data. Performance values reflect the approved selected 30-day scope.
Your workbook is the source of truth.
Upload .xlsx or .xlsb. We discover inputs and outputs automatically, preview the model, and surface cell-level validation errors before compiling — asynchronously, at any size — to deterministic bytecode: content-addressed, signed, cached.
- .xlsx + .xlsb · 380+ functions
- Automatic IO discovery
- Preview + actionable validation errors
- Reliable async compilation
Sheets
83
Cells
22.4M
Formulas
2.47M
Functions
380+
INPUTS · 130 named ranges
age: number · region: string · cover: number
OUTPUTS · 112 named ranges
premium: number · tier: enum · currency: string
Every upload is a new version. The old one keeps serving.
Push v5 while v4 handles live traffic. Pin any environment to any version. One-click rollback; a compile is a deploy.
- Immutable version IDs
- Pin prod / staging / dev separately
- Diff named outputs between versions
- Deterministic replay of any call
| Ver | Env / Tag | By | Age |
|---|---|---|---|
| v4 | prodDefault | risto.t | 2h ago |
| v3 | prodPrevious | elise.p | 5d ago |
| v2 | staging— | mihkel.t | 14d ago |
| v1 | —Archived | risto.t | 34d ago |
Approved quotes become regression tests.
Save approved quotes and calculations as test cases. Every new workbook version runs the full suite, and you see exactly which outputs changed — field by field — before production promotion.
- Capture cases from real calls
- Runs against every new version
- Field-level output diffs
- Gate promotion on a green run
Tokens scoped to model + version + env.
Generate API tokens with fine-grained scope. Revoke in one click. Rotate on a schedule. Every call is logged to the audit stream.
- Env: prod · staging · dev
- Per-token rate limits TBD
- Audit log retained 2 years
- SCIM + SSO on Enterprise
| Token | Env | Model | Status |
|---|---|---|---|
| ctkn_9af2_xQr12… | prod | rating_v4 | Scoped |
| ctkn_41bb_qnaH… | staging · latest | rating_* | Rotation due |
| ctkn_0071_bds9… | dev · any | * | Revoked |
Mock calls without shipping code.
Everyone on the team — not just engineers — tests what the API will return. Type inputs, see JSON. Copy to cURL / Python / TS. Save presets.
- Live against prod or any version
- Saved I/O presets
- Copy as cURL / Python / TS
- Sharable permalinks TBD
{ "age": 34, "region": "EE", "cover": 50000, "tier_hint": "STD" }
{ "premium": 287.42, "tier": "STANDARD", "currency": "EUR", "discount": 0.12 }
Understand and allocate every execution.
Volume, errors, success rates, and latency across models, customers, products, and destinations. Apply business labels to models and allocate usage by customer, product, brand, or business unit. Errors link back to the exact cell.
- Prometheus + OpenTelemetry exporters TBD
- Per-cell error resolution
- Labels: customer · product · brand · unit
- Usage by destination & model
Avg latency
2ms
Hot path
1ms
Cold path
89ms
Executions
7.5M
Success rate
99.6%
Usage reports, delivered into your S3.
Hourly or daily aggregate usage as versioned Parquet files, dropped directly into your bucket. Query with Athena; feed billing, reporting, or cost-allocation workflows. Exports contain allocation data—not calculation inputs, outputs, tokens, IP addresses, or pricing information.
- Hourly + daily aggregates
- Versioned Parquet · Athena-ready
- Billing & cost-allocation ready
- No sensitive calculation data
date=2026-08-09/hour=13/usage_a41f.v3.parquet 12.4 KB date=2026-08-09/hour=14/usage_a41f.v3.parquet 11.9 KB date=2026-08-09/usage_daily.v3.parquet 88.1 KB
athena> SELECT label_customer, SUM(executions) FROM eucalyptus_usage GROUP BY 1; acme_re 541,204 nordica 372,118 baltic_re 164,332
Run the execution plane in your own AWS.
The same Rust VM, deployed in your VPC as a private Destination — your account, your region. Calculation traffic, workbooks, compiled models, tokens, and execution history stay under your control.
- Any AWS region or account
- Private networking
- Bring your own KMS
- Scales to zero
# eucalyptus.cfn.yaml AWSTemplateFormatVersion: "2010-09-09" Description: Eucalyptus self-hosted Destination Resources: Runtime: Type: AWS::CloudFormation::Stack Properties: TemplateURL: https://cfn.eucalyptus.sh/runtime/0.7/aws.yaml Parameters: Region: eu-central-1 VpcId: !Ref ProdVpc KmsKeyArn: !GetAtt EuKey.Arn LambdaMem: 2048
OIDC SSO + SCIM 2.0, out of the box.
Connect your identity provider with OIDC SSO, map IdP groups to application roles, and automate user provisioning and deprovisioning through SCIM 2.0.
- Okta, Entra ID, any OIDC IdP
- IdP groups → application roles
- Automated provisioning + deprovisioning
- Access changes audited
Group → role mapping
idp:underwriting→ role: model-editor
idp:platform-eng→ role: admin
idp:finance→ role: usage-viewer
SCIM 2.0 · provisioning
+ elise.p provisioned · model-editor · 2m ago
− j.moss deprovisioned · left org · 1d ago
Built for people and AI agents.
Manage the model lifecycle from the dashboard, API, CI pipeline, or an OAuth-secured MCP agent. Agents can upload workbooks, discover IO, compile versions, run tests, and inspect logs and metrics through a least-privilege toolset.
- Works with any MCP client
- Upload · discover IO · compile · test
- Inspect logs + metrics
- OAuth-secured, least-privilege, audited
▸ tool upload_workbook("rating_v6.xlsb") ← version v6 · compile queued ▸ tool discover_io("rating", "v6") ← 130 inputs / 112 outputs · typed ▸ tool run_tests("rating", "v6") ✓ 23 unchanged ▲ 1 changed ▸ tool read_metrics("rating", { window: "24h" }) ← execs 7.5M · success 99.6% · avg 2ms▍
Try it