← All projects
Build 01AI executive intelligenceCore built · connectors next

RIN

Ask your company “why?” Get an answer you can trace.

ORION connects a company's scattered information, investigates business questions across it and explains what the evidence supports. When the evidence is thin, it says so.

RoleSolo · design to backend
ForGrowing companies with scattered tools
CoreAsk · Investigate · Workspaces
Proof68 tests · negative control
StackFastAPI · Postgres + pgvector · Next.js
01 · The problemWhy this exists

The answer is in six tools. Nobody can see all six.

Picture a 100-person company. Customer talks live in the CRM, money in spreadsheets, projects in Jira, decisions in Slack, docs in Drive, complaints in Zendesk. When the CEO asks a hard question, three people spend days stitching it together.

CRMSPREADSHEETSJIRA · LINEARSLACKDRIVE · NOTIONORIONONE BRIEF+ ZENDESK, FINANCE DB, …
The CEO asks
“Why did revenue drop last month, and what should we do about it?”
Finance analystPULLS THE NUMBERS
Customer successCHECKS ACCOUNTS
Operations leadDIGS THROUGH PROJECTS

ORION gives all three a much faster starting point, with every finding linked back to its source. It doesn't replace them. It does the gathering.

02 · The productThree surfaces

A dashboard, a research assistant and an investigator.

One product, three ways in. Leaders see what changed, ask questions in plain English, and open a full investigation when the answer isn't simple.

ORION · EXECUTIVE DASHBOARDCONCEPT UI · ILLUSTRATIVE DATA
NORTHWIND CO · JULYLAST 30 DAYS ▾
REVENUE$410K▼ 18% vs June
RENEWALS AT RISK3▲ 2 this month
OPEN ESCALATIONS12▲ 5 this week
PROJECTS ON TRACK9/112 delayed
REVENUE · LAST 6 MONTHS
JUL −18%FEBMARAPRMAYJUNJUL
ORION FOUND
● REVENUE
88.9% of the drop comes from one customer. Investigate →
● RENEWAL RISK
Cedar Ridge: usage reporting delayed, export issues unresolved.
● DELIVERY
2 projects slipped past their deadlines.

See what changed

Revenue, customer risk and delivery in one view, with ORION's findings beside the numbers instead of hidden in fifteen dashboards.

Which customers are most likely to leave next quarter, and why?ASK ⏎
ORION · ANSWER

Two accounts show strong warning signs. Cedar Ridge has unresolved dashboard export issues and delayed adoption reporting ahead of renewal[1][2]. Atlas Freight cut active seats after a support escalation[3].

  • A third account looks weaker, but the evidence isn't strong enough to call it at risk.
CONFIDENCE MODERATE
SOURCES · 3
1
Renewal record · Cedar RidgeCRM · RENEWAL IN 41 DAYS
2
Ticket: dashboard export failingSUPPORT · OPEN 19 DAYS
3
Seat count change · Atlas FreightBILLING · −30% SEATS

Ask in plain English

Answers come only from the company's own connected information, with a numbered source for every claim and a clear note when the evidence falls short.

Why did revenue decrease by 18% in July?INVESTIGATING
INVESTIGATION PLAN
01Find the change and comparison periodDONE
02Rank customers and products driving itDONE
03Pull contracts, incidents and projects, including the weeks beforeDONE
04Test explanations against the evidenceRUNNING
05Write the executive briefQUEUED
DRAFT BRIEF
WHAT
Revenue fell 18%, $500K → $410K.
DRIVER
One customer, 88.9% of the decline.
LIKELY WHY
Service incidents in June, before the drop. Validating…
OPEN
Was the contract change planned?

Investigate, don't just answer

ORION breaks a hard question into steps, runs real tools against the data, checks each explanation against evidence and states what's still uncertain. This is what makes it more than a document chatbot.

PRODUCT DIRECTION · DESIGNED FOR THIS PAGEREAL SCREENSHOTS COMING WITH THE UI
03 · Built vs. visionNo overselling

The foundation is real. Here's exactly what works.

The full vision is ambitious, so I keep a clear line between what runs today and what's next. The hardest part, the investigation engine, is already built and tested.

Built 06

Document ingestion & semantic searchLIVE
Evidence-backed Ask with citationsLIVE
Multi-step investigation engineLIVE · IMPROVING
Company workspaces & data isolationLIVE
Data source foundationLIVE
Idempotent data importsLIVE

Next 05

Real connectors: Slack, Drive, CRM, JiraPLANNED
Automatic ongoing syncPLANNED
Proactive risk monitoring & alertsPLANNED
Scheduled executive briefingsPLANNED
Production SaaS deployment & access controlFUTURE
68automated tests passing in CI on Postgres + pgvector
18%revenue decline found and computed in evaluation
88.9%of that decline traced to a single customer
0.44confidence cap when the evidence is removed
04 · The result I'm proudest ofNegative control

I tested whether it would refuse to guess.

Same question, same company, run twice. The second time I deleted the evidence that explains the drop. A tool that's guessing tells the same story both times. ORION didn't.

FIG. 01 · THE SAME QUESTION, TWICEFROM ORION'S TEST SUITE
A · With the evidence
Why did revenue decline in July?
WHAT HAPPENED
Revenue fell 18%, $500K → $410K.
DRIVER
One customer, 88.9% of the decline.
EXPLANATION
Validated against records from the weeks before.
CONFIDENCE
Validated
B · Evidence removed
Why did revenue decline in July?
WHAT HAPPENED
Revenue fell 18%, $500K → $410K.
DRIVER
One customer, 88.9% of the decline.
EXPLANATION
None stated. The evidence doesn't prove a cause.
CONFIDENCE
Capped · 0.44
Engineering note

Testing also caught a real planner bug: the AI only searched for evidence in the month revenue fell, missing incidents from the weeks before. The validator now always reaches back a month, whatever the model plans, and a regression test replays the exact failing plan.

05 · The visionSix capabilities

From answering questions to spotting problems first.

Today, most AI assistants wait to be asked. The finished ORION connects, investigates, watches and helps people act, with humans approving anything important.

AFOUNDATION BUILT

Connect

Link approved tools into one permission-aware map of documents, records, metrics and how they relate.

The point isn't collecting data. It's connecting what lives apart.
BBUILT

Ask

Ask hard business questions and get answers from the company's own evidence, with sources.

If the evidence isn't strong enough, ORION says so.
CBUILT

Investigate

Break a question into steps, pull evidence across departments, test explanations and report confidence.

Its biggest differentiator: it investigates, not just responds.
DPLANNED

Monitor

Watch approved sources for meaningful changes and flag risks before someone has to ask.

● RENEWAL RISKILLUSTRATIVE
Cedar Ridge · renewal at risk

Reporting delayed, export issues unresolved. Review with customer success.

EPLANNED

Brief

A short Monday-morning briefing: what changed, who needs attention, what's moving the wrong way, and the evidence for each.

Less time gathering. More time deciding.
FPLANNED

Act

Propose the next step, like a Jira task or a customer follow-up, and run it only after explicit approval.

Not an AI that runs the business. One that helps people act responsibly.
06 · How it worksTarget architecture

Built in layers, for decisions that cost money.

Some layers are implemented and others are planned. Every layer follows the same principles, because a wrong business answer has real financial consequences.

L1
Connected business systems

CRM · Documents · Projects · Support · Finance

PLANNED
L2
Ingestion & sync

Parsing · Normalization · Change detection · Permissions

PARTIAL
L3
Company knowledge layer

PostgreSQL · pgvector · Documents · Metrics · Relationships

BUILT
L4
Reasoning & investigation engine

Retrieval · Planning · Tool execution · Evidence validation

BUILT
Ask
Investigate
Insights · next
Design principles
01Isolate each company's data
02Keep the source of every piece of evidence
03Put hard limits on what the AI can do
04Test the failure cases, not just the happy path
05Make uncertainty visible
07 · Who it's forFirst market: growing companies
CEO / Founder

Company performance and emerging risks

COO

Operational bottlenecks and delayed projects

CFO / Finance

Revenue changes and financial trends

Customer success

Renewal risks and unresolved issues

Product / Eng

Linking customer problems to delivery

A GENERAL AI ASSISTANT
  • Knows the internet, not your company
  • Answers from what it read in training
  • One conversation at a time
  • Waits to be asked
ORION
  • Connected to the company's own systems
  • Every claim traced to a source, or withheld
  • Structured, multi-step investigations
  • Permission-aware and, next, proactive

Not “ORION knows everything about your business.” ORION helps you investigate it, with evidence you can trace.

FastAPISQLAlchemy 2PostgreSQL 16pgvectorRedisCeleryNext.js 15React 19TypeScriptOpenAIGitHub Actions
↑↓ NAVIGATE↵ OPENESC CLOSE