Component data lives in 40-page PDFs and is duplicated across every tool that needs it. Oriv unifies public & private data sheets, and supplier qualifications into one canonical schema, queryable from any tool your engineers already use. EDA, PLM, ERP, AI agents, internal apps. The data layer regulated hardware teams have been hand-stitching for twenty years.
// Cited. Traceable. Yours.
// 4 vendor formats → 1 canonical record
↓ ORIV ↓
supply_voltage_maximum: 5.5
unit: V (SI base)
condition: { Tj: 25°C }
provenance: page_3 / table_2.4
confidence: 0.997
// Plugs into your existing stack
EDA / CAD
PLM
ERP / DATA
AI & AGENTS
CUSTOM
One data layer. Every tool your engineers already run.
Every tool in your stack carries its own version of the same component data. The EDA team re-extracts the parameters. The PLM team updates them on a different cadence. The BOM team picks the part number it found first. AI agents come last and guess. The most visible example sits right below.
// Same prompt. One knows what the words mean.
“Find me an automotive-grade op-amp with VOS max under 100 µV. Show me the candidates and their footprints.”
> oriv.query op_amps {
vos_max < 100 µV
aec_q100 = true
}
Resolving "vos_max" against condition tuple:
Tj = -40 to +125 °C (AEC-Q100 range).
3 candidates indexed.
2 pass the over-temperature spec.
ADA4528-1 · Analog Devices
vos max (Tj=25°C) 2.5 µV p.3, tbl 2.1
vos max (-40..125°C) 25 µV p.4, tbl 2.2 ✓
aec_q100_grade 1 p.1
package SOIC-8 p.21, drawing
OPA388 · Texas Instruments
vos max (Tj=25°C) 5 µV p.6, tbl 6.5
vos max (-40..125°C) 90 µV p.7, tbl 6.6 ✓
aec_q100_grade 1 p.1
package SOT23-5 p.18, drawing
OPA320 dropped: product-page "40 µV" is typical
at 25°C. Over operating range, vos max = 200 µV.
Footprints, pinouts, IPC-7351B land patterns
exported for both. Altium · KiCad · Allegro.From web search, three strong AEC-Q100
candidates with low offset:
• ADA4528-1 (Analog Devices)
VOS = 2.5 µV. SOIC-8 package.
• OPA388 (TI)
VOS = 5 µV. SOT23-5 or SOIC-8.
• OPA320 (TI)
VOS = 40 µV. SOT23-5 standard.
All three are AEC-Q100 qualified. Any
meets your VOS requirement.
Footprints: I can describe these packages
generally (SOIC-8: 8 leads, 1.27 mm pitch;
SOT23-5: 5 leads, 0.95 mm pitch). For exact
IPC-7351B land-pattern geometry, consult
the datasheet or use your CAD tool's wizard.
Recommendation: ADA4528-1 for tightest spec.Same prompt. The LLM doesn't know which “VOS” is which, or how to generate a footprint. Oriv does.
The component databases your industry uses today index public datasheets only. Useful, but it leaves the regulated, certified, custom-spec part of your BOM untouched. That's the part your engineers actually fight with. Oriv ships both layers in one queryable schema.
SiliconExpert, Z2Data, Octopart, Findchips, Accuris. Twenty years of indexing the headline parameters of every public datasheet. Supply range, package code, qualified grade. The condition tuples, derating curves, and footnote constraints behind them aren't in their schema. Your private SCDs and supplier qualifications aren't either.
That's the wedge. We extract the entire datasheet as parametric, semantic data: every condition, every curve, every constraint, with provenance. Plus your private documents in the same schema. In your tenant.
Your data never leaves. No cross-customer training. The schema is the moat, not your data.
// Typical BOM coverage · Regulated industry
Public corpus
Vendor datasheets, lifecycle status, distributor inventory.
Your private layer
Custom SCDs, MIL-PRF, supplier qualifications, NDA releases.
Not a flat catalog. Not a chat interface. A real parametric store with canonical schemas, validated units, preserved conditions, and per-field provenance. Queryable through standard SQL, MQL, SPARQL, or directly from any application in your stack.
Vcc, VDD, “Supply Voltage”, “VS+” all collapse to the same canonical field. Public corpus and private layer alike.
// fixed schema, growing
Every numeric value reduced to SI base units, dimensionally validated. mV, µA, kHz, °F all converge. No silent unit drift.
// SI base · validated
“100 mA at 25°C, 60 mA at 85°C” extracts as a structured tuple, not a string. Operating conditions are first-class data.
// {temp, supply, load}
Every record carries a citation back to source page, table, and footnote. Auditable for ISO 26262, FDA, DMSMS workflows.
// page · table · footnote
SQL, MQL, or SPARQL on a real parametric store. Real columns, real joins, no hallucination. Plus direct access from AI agents.
// SQL · MQL · SPARQL · agent-native
Your private extensions never leave your tenant. No cross-customer training. The public library is shared. Your long tail is yours.
// VPC · on-prem available
Oriv ships the data layer. The applications below are customer-built, often with Cursor or Claude Code doing most of the typing. Names redacted under NDA. Patterns are real.
Pattern from A&D conversations. Today's DMSMS investigation runs 4 to 6 weeks per part. With Oriv as the data layer, the same investigation runs in a day. The team keeps their existing tooling, Oriv ships the data underneath.
// 6 weeks → 1 day per investigation
For:DMSMS Office
Replaces:Excel + tribal knowledge
Pattern from hardware R&D conversations. Engineers spend a day per component triaging datasheets. With Oriv connected to their AI agent, asking for an alternate to IRF7314 with lower RDS(on) returns cited candidates in seconds. With checked Pin compatibility.
// 200 candidates → 8 cited alternates in 90s
For:Hardware R&D
Replaces:Manual datasheet triage
Pattern from avionics test conversations. Setting up a new HIL bench takes two weeks of LabVIEW and Python wiring per rig. With Oriv as the data layer, an AI agent understands the system, sets up the right widgets, and generates the glue logic. Same morning, working test setup.
// New HIL rig: 2 weeks → 1 morning
For:Test Engineering
Replaces:Per-rig custom integration code
Pattern from robotics conversations. Picking a new IMU or ToF sensor means cross-checking datasheets for drift, interface, and stock. With Oriv, the agent answers "find me an IMU with tighter drift and the same I2C address" in seconds. Cited against prior-qualified parts.
// Sensor swap: 2 days → 20 minutes
For:Robotics R&D
Replaces:Cross-checking by hand
Pattern from robotics prototyping conversations. Building a calibrated simulation model from a datasheet runs 2 to 3 days per component. With Oriv as the data layer, an AI agent pulls conditions, signal characteristics, and qualification limits, and emits a ready model.
// Behavioral model: 3 days → 4 hours
For:Prototyping
Replaces:Manual model calibration
Pattern from hardware R&D conversations. Five MOSFET candidates, five datasheets, five different ways to spec "RDS(on)". Engineers build the comparison spreadsheet by hand. With Oriv, one query returns all five at the same operating point.
// Cross-vendor comparison: half a day → 30 seconds
For:Component Engineering
Replaces:Manual cross-vendor spreadsheets
We're picking three teams for the Q3 cohort, one per category vertical. Co-shaped canonical schema for your domain. Direct founder line. Locked-in pricing for three years post-GA.
// Design partner gets