Products / TERUSAN
S.A.L.T CI TERUSAN
Governed process flow. Any approval, application or case is written once as a blueprint; people and AI then work inside it, and every step is recorded. Directed, governed and visible.
Done · 2 d
Done · 3 d
In review
AI check · level 2 2 supporting items are missing: the cost breakdown and the vendor letter. Add them, then resubmit.
09:15 AI check drafted
10:02 Bursar approved
10:03 Routed to risk review
Processes that live in documents lose time at every hand-off
Most approval processes run on Word papers, spreadsheets and e-mail. Details are retyped, totals are copied by hand, and nobody can say where a case is without a phone call.
Sections typed separately across three papers in one real approval chain
Supporting items in a checklist ticked by hand, per project category
In scoring formulas of a real evaluation workbook
Totals carried by hand from one form to the next
From an analysis of a real six-stage ICT project approval chain at a public institution.
What the documents tell us
Type it once
Project details recur across papers. They should flow from one record into every document.
Calculate it, do not copy it
Weighted scores and totals belong in the system, not in hand-built formulas.
Show where it is
Six hand-offs between roles. Anyone should see where a case stands and who holds it.
TERUSAN in one line
Terusan is Malay for a canal: water that goes where it is directed. TERUSAN is the process layer of S.A.L.T CI, one governed engine that runs any long-lived, multi-person process, with AI working inside the rules.
One governed engine for any process
A process is written once as a blueprint. The blueprint fixes the allowed states, roles, rules and deadlines; people and AI work inside it; every action is recorded.
Local first
Engine, models and data on the organisation's own server; works with the network cut.
People decide
Every consequential step passes a person, and the advice they saw is logged.
Configuration over code
Order, thresholds, forms, weights and people are blueprint data, never a code change.
Everything is an event
One log feeds the audit trail, the case timeline and the analytics.
The process is a contract. AI works inside it.
Agents can draft, check, summarise and nudge. They cannot skip a step, change a rule or approve on a person's behalf.
A clear view for every role
Applicants, committees, management and process owners each see only what they need, in Bahasa Melayu or English, on a computer or a phone.
Where is my application?
- Start an application and fill papers from one record
- Stage bar with days taken per stage
- AI completeness check before submission
- Answered on screen, without a phone call
Decide with the evidence in front of you
- Review, approve, return or reject with comments
- Panel members score independently; totals calculate themselves
- Sign-off panels with ordered signatures
- Advice, sources and model logged with each decision
Where are we stuck?
- Open, overdue and average days per stage
- Bottlenecks and forecasts with the evidence behind them
- Advice you can trace back to its data
Change the process without calling the vendor
- Edit deadlines, weights, forms and routing
- Draft, simulate, then publish
- Every change versioned and audited
Thirteen blocks, any process
Long-lived, multi-person processes are combinations of the same few kinds of step. A new process is a new combination; engine code changes only when a new kind of block is needed.
Ask and decide
- Human task
- Review
- Sign-off panel
Check and score
- AI step
- Rule table
- Scoring
- Compute
Make and send
- Generate document
- Tool call
- Notify
Route and wait
- Wait or timer
- Branch, parallel, join
- Sub-process
Processes the same engine runs
| Process | How the blocks combine |
|---|---|
| ICT project approval chain | Papers, reviews, a required-documents rule table, weighted tender scoring and a six-signature sign-off |
| Leave or claim approval | The smallest useful blueprint: one form, one review, one notification |
| Small purchase with quotations | The same scoring model with different weights |
| Staff recruitment | A panel scores independently, then results merge |
| Research or proposal evaluation | Many proposals ranked against a fixed budget |
| Meeting to action | Builds on MOM Studio minutes: each action becomes a tracked task |
| Security alert to incident | PERISAI alerts start a governed response |
| Complaint or service request | Measured by time to resolution, with escalation timers |
AI that is useful, bounded and traceable
Each AI step in a blueprint has its own autonomy level. A process owner starts low, measures accuracy on real documents, and raises the level only when the evidence supports it.
No AI on this step.
Shown beside the work; the person types the result. e.g. a suggested score.
Pre-filled; the person reviews and confirms. e.g. a summary of a bidder's reply.
Applied; visible, revertible and sampled. e.g. a total read from a quotation.
Applied and logged. e.g. reformatting a date field.
The evidence rule
Every AI output is stored with the model, the instruction version, its inputs, the sources it cited and its confidence. Low-confidence or uncited output goes to a person, never straight through.
Three agents, each with a narrow job
Drafts blueprints
Turns an uploaded form or a described process into a draft blueprint for the process owner to review, simulate and publish.
Helps people in a case
Checks completeness, drafts summaries and explains what is missing, inside the step it is attached to.
Keeps cases moving
Sends reminders, escalates overdue steps and routes cases, only along paths the blueprint allows.
Every option is tunable. Nothing is hard-wired.
Departments differ. TERUSAN keeps one base process and lets each department adjust the parts that are theirs, with the right person holding each control.
| Layer | What changes | Who holds it |
|---|---|---|
| 1 · Knobs | Deadlines, reminder timing, score weights, amount thresholds, committee members | Department administrator |
| 2 · Rules | Routing conditions, required-document matrices, escalation rules | Process owner |
| 3 · Structure | Stages and their order, parallel or conditional branches, paper sections and fields, scoring criteria | Process owner |
| 4 · AI behaviour | Per step: autonomy level, instructions, model, knowledge sources, output format | Process owner with platform administrator |
| 5 · New blocks | A step type or tool that does not exist yet | S.A.L.T CI engineering |
One base process, many departments
instead of 20 / 80
for its own sign-off
instead of 3
Versioned blueprints
A running case keeps the version it started with; new cases use the new one.
Draft, simulate, publish
Run a change against sample cases before anyone relies on it.
Audited changes
Who changed what, when and why, in the same log as the cases.
Dashboards that answer questions. Advice you can defend.
Because every action is an event, management gets its answers from the same log that provides the audit trail. Nothing is re-keyed into a report.
Dashboard
What is open, where it is stuck, what is overdue, and how work is spread across roles.
Process analysis
Where time goes per stage, per department and per blueprint version.
Anomalies and duplicates
Does this case look unusual, or like one already seen?
Advice per case
What the criteria, the cited policy and comparable past cases suggest, and the risks.
Portfolio advice
Given a fixed budget, how open items rank on the chosen criteria.
Forecast
When a case is likely to finish, given how similar ones went.
Guardrails on advice
Advice respects permissions, is about the process and not about individuals, says so when data is thin, and can be reproduced from the same log.
A record that cannot be quietly changed
Every action is written to the case history and cannot be edited afterwards. Documents are generated from the case, not retyped.
Hash-chained history
An append-only event log chained by hash, so any alteration or deletion shows. Administrators can verify the chain at any time.
Generated papers
Word documents produced at the end of each stage from the case data, in the organisation's own form layouts.
Audit export
The full event log of a case downloads as CSV for auditors.
Need-to-see access
Applicants see their own cases. Reviewers see a case while it waits for them, or after they acted. Management sees all.
Frozen settings
Each case records the settings it runs on, so a later change never rewrites history.
Notices and reminders
A bell for returns and decisions, and reminders as deadlines approach or pass.
Example chain on TERUSAN
Paper · review
Paper · review
Rule table · AI check
Paper · compute
Scoring · compute
Paper · signature panel
Section lists, the required-documents matrix, score weights, signatory roles and Malay labels are blueprint data, not product code.
Inside the organisation's own network, which can be disconnected
Case data, documents, the event log and knowledge sources stay on the organisation's server. External AI is an optional connector, off by default.
Browser or phone
Process engine
The only route to AI
On the organisation's server
Optional; enabled per step by an administrator
Local models by default
Every AI call goes through the S.A.L.T CI gateway, which uses local models first.
No secrets in blueprints
Keys stay in the deployment's secret store and are referenced by name; AI keys are stored encrypted.
Access you control
Role-based access and single sign-on with the organisation's directory.
One server to start
A single container deployment; a second node for redundancy is optional.
Visible and recoverable
Documented install, backup, restore and upgrade; queue views show any stuck job.
Two languages
Bahasa Melayu by default with English at a click, on desktop and phone layouts.
Five questions to ask any process-automation vendor
Process automation is a crowded field, and the answers matter more than the demo.
| Question | TERUSAN answer |
|---|---|
| Where do the documents and AI models run? | On the organisation's own server. Local models by default, and it keeps working with the network cut. |
| Who changes the process: you or the vendor? | The organisation's process owners, from a designer, with versions, simulation and an audit trail. |
| Can every AI suggestion be traced? | Yes: model, instruction version, inputs, cited sources and confidence. Weak or uncited output goes to a person. |
| Does it work in Bahasa Melayu and with our own forms? | Roles, form codes and labels live in the blueprint, in Malay or English; AI steps are benchmarked on the organisation's samples first. |
| Is it one more silo, or part of a platform? | Part of S.A.L.T CI: the same gateway, identity and shared services as MOM Studio and the rest. |
From blueprint to a running process
Read the real forms and hand-offs
Stages, roles, rules and papers
Run sample cases end to end
One department, real users
Variants for other departments
How to compare fairly
Give each vendor the same real process, ask them to run it on your own server, then cut the network. S.A.L.T CI is happy to be measured that way.
Part of one sovereign platform
TERUSAN runs on the TERAS core with the same sign-in, audit and models as every other S.A.L.T CI application.