The Company of Truth
Decide on what is true.
Satsang AI builds decision intelligence for operations where a wrong call is expensive. We find what is true in what your organisation already knows, show you the evidence behind it, and leave the decision with your people.
The Company of Truth
Satsang (n.) · being in the company of truth
Customers make better decisions when they can trust what they know.
So everything we do answers to one test: is it true, and can you see why?
That test applies to the problem we choose to solve, the data we build on, the product your people use, the way we work with you, and what we promise at the start. The five truths below are how we hold ourselves to it. Each one comes with the mechanism that makes it more than a slogan.
For Decision owners
Skip to 02 · In the dataTruth 01In the problem
Before we build anything, we find the real problem.
In short: We start with the people who make the decision, not the technology. The knowledge usually exists. It just isn't connected.
Where we start
Three steps, in this order, before anything is built.
Start with the customer
Who makes the decision, what it costs to get wrong, and what they reach for today.
Name one decision
One decision that matters, narrow enough to finish and real enough to measure.
Test it on real cases
Past cases with known answers tell us whether the problem is the one we think it is.
What we usually find
And what the same work looks like once the knowledge is connected.
| Area | What we usually find | Connected |
|---|---|---|
| Cases | Each failure is investigated from scratch | Similar past cases surface with their sources |
| Similar assets | A fix learned in one place stays in one inbox | One lesson reaches every like-for-like asset |
| Findings | Inspections and audits sit in separate systems | One view of findings per asset and per equipment type |
| Decisions | Options built from memory, calls and email threads | Options with evidence, risk and impact side by side |
| People | Expertise leaves when the expert leaves | Every closed case adds to the organisation's memory |
Cases
Each failure is investigated from scratch
Similar past cases surface with their sources
Similar assets
A fix learned in one place stays in one inbox
One lesson reaches every like-for-like asset
Findings
Inspections and audits sit in separate systems
One view of findings per asset and per equipment type
Decisions
Options built from memory, calls and email threads
Options with evidence, risk and impact side by side
People
Expertise leaves when the expert leaves
Every closed case adds to the organisation's memory
Automating paperwork saves minutes. Getting the decision right saves far more.
For Data and IT leads
Skip to 03 · In the productTruth 02In the data
Establish what is true. Then find its story.
In short: Records say what happened. Reports and past decisions say why. We connect both, so every answer rests on something you can check.
One loop, not four tools
Every closed case returns as knowledge, so the foundation gets better with use.
Connect the data
What happened: records, logs, histories, costs.
Capture the knowledge
Why it happened: reports, manuals, findings, the reasons behind past decisions.
Close the loop
What you learn: each decision and its outcome return as knowledge.
Deliver the experience
How it is used: apps and agents inside the working day.
Connect the data
What happened: records, logs, histories, costs.
Capture the knowledge
Why it happened: reports, manuals, findings, the reasons behind past decisions.
Deliver the experience
How it is used: apps and agents inside the working day.
Close the loop
What you learn: each decision and its outcome return as knowledge.
The next case starts smarter
How the data stays true
An agent reasons only as well as the data under it. Each property below comes with the rule that enforces it.
Discoverable
Every source is registered before it is read.
Its owner, scope, approval and freshness are recorded first, so nothing enters unaccounted for.
Connected
One question reaches every related record.
Records are tied to the asset or equipment type they are about. Where a link cannot be made it is left empty and counted, never guessed.
Unaltered
Originals are never changed.
Each original is stored once, under its own content hash. Indexes and summaries can be thrown away and rebuilt from it.
Defined
Every number means one thing.
Each metric has one definition, one owner and a version. What happened is kept apart from notes on how to work.
Four guarantees
- Nothing you send is altered.
- Every answer traces to a source.
- Gaps are counted, never guessed.
- Every decision can be reconstructed later.
For Everyday users
Skip to 04 · In how we workTruth 03In the product
Understand what is happening. Decide what to do. Act.
In short: The agent gathers. Your people decide. Every step shows its working, so you see why an option is there before choosing it.
One case, start to finish
Understand
What is happening, and what has happened before.
Arrives
Report
A report or a request comes in.
Agent
Connect
The agent links it to the asset, its history and its open items.
Agent
Recall
It gathers similar cases, guidance and findings, each with its source.
Decide
Options with evidence, risk and impact side by side.
Person
Decide
Options arrive with evidence. A person makes the call.
Act
Approved action, and an outcome that teaches the system.
PersonAgent drafts
Act and learn
A person approves the drafts. The verified outcome becomes knowledge.
Every answer shows its working
Each promise is enforced by the system, not left to good intentions.
Cited
Every answer shows its sources.
A claim is kept only if a tool returned its evidence in that same run. Unsupported claims are removed, not repaired.
A person decides
Nothing is sent and nothing is raised without approval.
The agent can prepare a case and stop. Deciding, approving and closing belong to a named person, and the server enforces it.
Permission-aware
People see only what the agreed access policy allows.
The access rule is part of the query itself, so a ranking or a count cannot leak a hidden record. The agent reads under the same rule.
Auditable
Question, sources, options and decision are all recorded.
An append-only record of what the agent retrieved, what it showed and what the person chose. A decision can be walked back to the original file it came from.
Honest about doubt
A suggestion is labelled, never pre-selected.
It is withheld when evidence is thin, and each option shows what would change it. High agreement with falling decision time is treated as a warning, not a win.
- Trust is what gets it used. Use is what makes it learn.
For Project sponsors
Skip to 05 · In what we promiseTruth 04In how we work
We ask the hard questions early.
In short: Being useful means saying what nobody wants to hear, including about our own work. Then twelve weeks and a plain answer.
Five practices
Each one is paired with the mechanism that makes it true.
We ask hard questions.
A baseline is measured before any target is set, so we know what “better” has to beat.
We challenge ideas, including ours.
Fixed cases with known answers decide what ships. Opinion, ours included, does not.
We surface problems early.
Missing links, numbers and sources are counted and shown at every review, not discovered at the end.
We learn from mistakes.
Every override and correction is recorded, and the lessons a person confirms feed the next case.
We change when the evidence changes.
Three reviews, about a month apart. Each one ends in a plain call: continue, change or stop.
Twelve weeks, then a plain answer
One decision that matters, on real cases, with the people who make it, beside your named owner.
Connect
Access agreed and scoped. A baseline is measured before anything changes.
Know
Knowledge connected and searchable. A first working version in people's hands.
Decide
Decision support on live cases. The people using it rate every output.
Prove
Results read against the baseline. Continue, change or stop.
- Continue, change or stop
What we ask of you
Truth runs both ways. Each new use case then reuses the same data, knowledge and case memory.
Scope
One area, narrow enough to finish.
Owner
A named person who owns the result, not only the project.
Design partners
Two or three practitioners, a few hours a week, to use it, rate it and agree past cases with known answers.
Access
Read access to the sources in scope, and a person who knows your codes. Approvals start before week 0.
Measures
A baseline measured first, then targets agreed against it.
Candour
Tell us when it is not working. We will do the same.
For Whoever signs off
Skip to How it is builtTruth 05In what we promise
We do not oversell.
In short: Success is your people using what we build on live cases and getting value from it, measured against where you started.
How we will know it worked
Three measures, read by you, not by us.
- Used.
Your people reach for it on live cases, without being asked to.
- Rated.
The people who use it rate every output, and the ratings hold up.
- Better.
Decisions improve against the baseline we measured together at the start.
Our commitments
Written down at the start, so you can hold us to them.
We will always tell you
- Where the work stands at every review: what is running and what comes next.
- What a result can and cannot show, before the result arrives.
- Where a link, a number or a source is missing. Gaps are counted, not filled in.
- When we think something should stop. Including our own work.
We will never
- Let an agent act without a person's approval.
- Promise a target before a baseline is measured.
- Claim a result we have not measured.
- Carry what we learn from your work to another client without a contract that says so.
Start with the customer. Find the truth. Build what matters.
For Technical teams
Skip to Start a conversationFor technical teams
How it is built, for the people who will check.
In short: Four layers and three functions that cross all of them. A request leaves only as an action a person approved.
Layer 1
Integration
Read first. Write back later.
- Everything that enters is read-only, within a scope you set.
- Every action that leaves has been approved by a person.
- Exports are enough to start. Interfaces follow once the value is proven.
- Your existing systems stay as they are.
Read first. Write back later.
- Everything that enters is read-only, within a scope you set.
- Every action that leaves has been approved by a person.
- Exports are enough to start. Interfaces follow once the value is proven.
- Your existing systems stay as they are.
One core owns the state, so policy and audit are written once.
Bring us one decision that matters.
Tell us where your experts spend their time gathering instead of deciding. We will tell you the truth about whether we can help, even when the answer is no.
Start a conversation