The workflow engine
for energy operations.
Not another SaaS dashboard. A programmable pipeline that turns documents into structured data, applies your rules, and produces auditable outputs.
Not just document AI.
Domain expertise built in.
7 years building for community solar. 30+ utility formats catalogued. 96.4% straight-through processing on production data.
General document AI can extract text from PDFs. What it can't do is know that a "community solar credit" isn't a "payment received." That a mid-cycle rate change needs special handling. That subscriber class A has different allocation rules than class B.
We built that knowledge into the extraction models themselves—not as prompts, but as training data from real operational workflows.
| General Document AI | Gridline | |
|---|---|---|
| Training approach | General-purpose + prompting | Domain-specific training |
| Energy terminology | Learned from prompts | Native understanding |
| Edge case handling | Requires custom rules | Built into model |
| Utility format coverage | Build per customer | 30+ pre-built |
| Human review needed | Frequent | Exception-based only |
| New format adaptation | Days to weeks | Hours |
What it replaces
The tools you're currently stitching together
Gridline replaces the fragile workflows that energy operators rely on. No more duct-taped spreadsheets, no more copy-paste from PDFs, no more wondering if a statement went out. Instead, you get a deterministic pipeline that handles document ingestion, data extraction, rule application, and output generation — with full audit trails and the ability to rerun any step.
allocate_by: "kwh_usage" fallback: "even_split" validate: required_fields How it works
Connectors → Workflow Engine → Templates → Data Model
Each use case is just a config + a few step functions, not a new app
The Pipeline
Six stages from document to delivery
These six primitives are the building blocks Gridline reuses for every workflow — billing in production today, interconnection and project finance next. Unlike typical document automation tools that use probabilistic ML models, Gridline combines hybrid NLP primitives: LLM-powered extraction with structured validation, rule-based computation with version control, and templated generation with full lineage tracking.
No other platform in energy operations combines deterministic extraction, declarative rule engines, and workflow-as-data in a single pipeline. This isn't RPA. This isn't generic ETL. It's purpose-built infrastructure for regulated industries handling financial documents.
Upload
PDF ingestion, file validation, S3 storage
bill.pdf Parse
LLM extraction, schema validation, confidence scoring
parsed_data.json Compute
Allocation rules, rate schedules, credit distribution
allocations.json Generate
PDF templating, statement rendering, branding
statement.pdf Invoice
PayPal sync, payment links, webhook handlers
invoice_id Deliver
Email dispatch, portal tokens, delivery confirmation
delivered: true Tech Stack
Production infrastructure, not prototypes
Where We're Pushing
The unrefined edges. The things nobody else is building.
Guided Workflow Builder
ShippedSeven guided steps from blank page to published workflow — no code, no JSON. Typed actions, searchable variable picker, publish blockers written as corrective actions.
Publishing is atomic: an immutable version with a sha256 plan hash. History compares, clones, and retires cleanly.
Deterministic Extraction
ShippedLLM-powered PDF parsing with structured output validation. Not probabilistic guessing — deterministic field extraction with confidence scores and fallback rules.
Most document AI is black-box. We expose confidence scores, let you set thresholds, and provide human-in-the-loop for low-confidence extractions.
Workflow-as-Data
ShippedRuns are first-class data objects, not ephemeral processes. Every step, attempt, and decision is queryable, replayable, and auditable — pinned to the exact version it ran.
Traditional ETL pipelines are fire-and-forget. We treat workflow execution as structured data you can inspect, debug, and replay.
Self-Serve Write Actions
BuildingBuilder-authored provider writes with preview, idempotency, and reconciliation built in — plus in-builder approval steps and browser authoring.
Each capability ships only after its security and reconciliation gate passes. The catalog labels everything native, managed, or planned — nothing planned is ever sold as available.
Security
Built for regulated industries handling financial data
Identity & Tenancy
Organization memberships and customer access grants; row-level security scopes every table to the tenant.
Credential Encryption
AES-256-GCM envelopes with key IDs, tamper detection, rotation. Secrets never in configs, logs, outputs, or screenshots.
Authorization
Server-side checks on every API and re-validation in the worker before each step — database, API, and runtime all enforce scope.
Runtime Egress Control
Outbound requests bounded: pinned hosts, TLS, same-origin redirects, DNS re-checks, private networks and metadata endpoints blocked.
Audit & Impersonation
Every sensitive action is an audit event. Support impersonation is read-only, 15-minute signed, fully logged.
Residency & Compliance
Region pinned per deployment — EU region available for Swiss / EU clients. DPA available; GDPR and Swiss revFADP aligned. Customer data is never used for training.
Runs & Artifacts
Everything is data. Everything is replayable.
Nothing mutates history
The pinning chain — every run executes exactly what was published
Editing never rewrites
Changing a workflow creates a new draft and version. In-flight and historical runs keep executing exactly what they pinned.
Rotation never breaks
Credentials float to the current secret; the contract stays pinned. Rotating a password strands nothing.
Evidence is audit-grade
Approvals record a hash of the exact data reviewed. Failed attempts, retries, and reruns are all preserved — never overwritten.
See it run.
No sales pitch. Get real access to test the product yourself on sample data.