Talk to an architectBook a 30-minute call
How it works

Checks first, AI second. Hosted in Dhaka.

How our agents decide, where they run, and what stays yours. Your records and rules decide first, AI only fills the gaps, and anything an agent can’t prove goes to a named person on your team.

Book a 30-minute call Talk to an architect
How it fits together · Your organisationHosted in Dhaka
What your people use
  • Agents
  • Your apps
  • AI workbenchEarly access
AI providers
  • Claude
  • GPT
  • Gemini
  • Open model in Dhaka
Company memory
  • Policies
  • Decisions
  • Past cases
Workflow engine
  • Steps
  • Rules
  • Hand-offs to a person
Your data layer
  • ERP
  • Documents
  • Microsoft and Google
Hosting
  • Dhaka · Tier-III
  • Your own servers
  • HyperscalerIf you choose

Company memory sits apart from the AI provider. Providers can change; what your company knows stays yours.

Checks first, AI second

Your records and rules decide. AI only fills the gaps.

Arithmetic, lookups and named checks run first. AI is called only where judgment is needed, and often only to pick from a shortlist. Anything still unproven goes to a named person on your team, with the reason. The result is fewer AI calls, a smaller bill, and no silent wrong answers.

Example · Vendor access requestsIllustrative
  1. 1
    Rules
    Same tax number and trade licence as an approved vendor.
    412
  2. 2
    Lookup
    Name or alias found in your vendor records.
    61
  3. 3
    AI picks from a shortlist
    Only on a miss. Chooses among three candidates; cannot invent one.
    19
  4. A person
    Anything still unproven, with the evidence attached.
    8
In this example batch
19 of 500
requests needed AI at all
0
approvals made by AI

AI can rescue a wrong rejection by picking the right match from the shortlist. It cannot approve anything. Numbers are illustrative.

Trade documents · Technolvelty

Verification that auto-approves only what it can prove. Arithmetic checks and named rules decide, an AI’s confidence score never approves, scanned text is checked rather than trusted, and pages are sorted without AI.

Read the story

Vendor access approval

Matching against your records runs first. AI runs only on a miss, and only picks from a shortlist, so it can rescue a wrong rejection but never create a false approval.

How it runs
  1. 01ReadDocuments and requests arrive.
  2. 02CheckArithmetic and named rules run first.
  3. 03MatchLookups against your own records.
  4. 04ShortlistOn a miss, AI picks from options.
  5. A personClears anything unproven, with evidence.
  6. 06RecordEach decision logged with its proof.
Stays in Bangladesh

Hosted in Dhaka. You decide what leaves.

Your apps, agents, databases, company memory, files and backups are hosted in a Tier-III data centre in Dhaka, or on your own servers if you prefer. Where each workload runs is set one by one. Outside AI providers stay off unless your policy allows them. Billing is in Bangladesh.

Where your workloads run · Example bankPolicy applied
Dhaka · Tier-III data centre
Applications
Agents
Databases
Company memory
Files
Backups
Your policy decides what goes out
Outside Bangladesh
Outside AI providers
Off unless your policy allows them
Off
Open model in Dhaka
Runs in country
In country

Set per workload

Customer records can stay in Dhaka while a public-data task runs elsewhere. One rule for everything is rarely the right rule.

Outside AI is opt-in

Calls to outside AI providers are off by default, allowed by policy for named tasks, and recorded when they happen. When allowed, the question and the context it needs are sent.

Billed in Bangladesh

Invoiced in Bangladesh and supported from Dhaka, with records ready for your auditors.

Not locked in: what can move
  • Your code
  • Your databases
  • Your data
  • Your way of working

Already on AWS, Azure or Google Cloud? Moving is a migration you plan, not a rebuild you fund. Run old and new side by side, check the results match, then switch the old one off.

Company memory

Your company’s knowledge stays with you. Change AI providers any time.

What your organisation knows (policies, decisions, past cases) is kept in a company memory in your own environment, apart from the AI. The platform is designed so you can use Claude, GPT, Gemini or an open model hosted in Dhaka, and move between them without rebuilding what your agents know.

Company memory and AI · kept apartMemory unchanged
AI provider · can change
ClaudeGPTGeminiOpen model in Dhaka
Company memory · yours, in your environment
Credit limit rule v3replaces v2 · approved by Risk
Current
Credit limit rule v2kept for audit · linked to v3
Replaced
New field: collateral typechange proposed
Awaiting approval
Your data and systems
Vendor and customer recordsDocumentsERPEmail

Knowledge stays with you

Policies, decisions, past cases and customer context are kept in your company memory, in your environment, not inside an AI provider’s account.

Choose the AI per job

A low-cost model for routine drafting, a stronger one for hard questions, an open model in Dhaka for sensitive text. Your policy decides.

Know what replaced what

Company memory records which decision replaced which, so an agent doesn’t answer from a rule that has been retired.

How it runs
  1. 01CaptureDecisions, policies and cases, with sources.
  2. 02LinkRecord which decision replaced which.
  3. A personApproves changes to rules and structure.
  4. 04RecallAgents answer from memory, citing sources.
  5. 05ChoosePick the AI per task, within policy.
  6. 06KeepYour knowledge stays yours.
What it runs on

Built, run and supported by Codemen.

Built on Groundfloor, an enterprise AI platform; Codemen is its implementation partner in Bangladesh.

You sign with Codemen. We design and build your agents, put them into production and keep them running. The platform handles the parts every app needs (sign-in, permissions, secrets, audit and deploys), so your budget goes on the work itself.

  • Workflow engine: the steps, rules and hand-offs to a person
  • Your data layer: an agent sees only what the person it works for can see
  • Hosting: a Tier-III data centre in Dhaka, your own servers, or a hyperscaler
Release checklist · Vendor portalHosted in Dhaka
Sign-in and SSO
Your company accounts
Permissions by role
Configured, not coded
Secrets
Stored outside the code
Database
Provisioned and backed up
Audit log
On for every action
AI access
Your policy applied
Exit and ownership

What’s yours stays yours. Even if you leave.

A fair question from any risk team: what if you, or the platform, go away? Here is the answer. The exact terms are set out in your contract.

Your data

You own it, and you can export it.

Your company memory

You own what your agents know, and you can export it.

Your agent logic

The agent logic we build for you is yours.

AI workbench Early access

Chat, plan, build, test and publish a small internal app in one conversation.

Your team describes what it needs; the workbench drafts it, and Codemen reviews it before it goes live. Apps run on the same platform as your agents, hosted in Dhaka.

Ask about early access
AI workbench · Leave approvalsEarly access
Build a leave-approval app for our Dhaka office.
WorkbenchPlan: three screens, approval by line manager, balance check against HR records.
PlanBuildTestPublished, hosted in Dhaka

AI Subscriptions

Claude, GPT and Gemini inside your AI workbench, with your company memory kept apart.

Reached through each provider’s API and billed in Bangladesh, with no separate ChatGPT or Claude accounts. Your policy decides which tasks may use an outside provider and which stay on an open model hosted in Dhaka.

See AI Subscriptions
AI access · By taskPolicy applied
Open model in Dhaka
Customer records, sensitive text
In country
Claude, GPT or Gemini
Public data, drafting, research
Outside, by policy
Company memory
Stays in your environment
Yours
Platform results

The platform has already shipped work.

21
app types delivered on the platform
155k
reference rows in a live master data model
99%+
processing accuracy on Westfield’s tenant sales data

Groundfloor platform results, not Codemen guarantees.

Client proof

In Bangladesh, and beyond.

Fiber@HomeBangladesh

Contract management for a national telecom network.

Contracts, renewals and obligations tracked in one place for a nationwide telecom infrastructure provider in Bangladesh.

Telecom infrastructureRead story
Pico Public CloudBangladesh

Groundfloor, live in Dhaka.

Our sovereign cloud partner. Groundfloor runs on Pico’s Tier-III data centre in Dhaka, so apps, agents and data can stay in Bangladesh.

CloudRead story
TechnolveltyInternational

Trade documents approved only when the numbers prove it.

Document verification for trade finance. Arithmetic checks and named rules decide; a model’s confidence score never approves. Anything unproven goes to a person.

Trade financeRead story

Names and screens are illustrative.

FAQ

Questions buyers ask

Something else? Ask an architect

Whose platform is this?
Codemen’s agents are built on Groundfloor, an enterprise AI platform. Codemen is its implementation partner in Bangladesh. You sign with Codemen, and Codemen designs, builds, runs and supports your agents.
What happens if Groundfloor goes away or changes its prices?
You own your data and your company memory, and can export them. The agent logic we build for you is yours. The exact terms are set out in your contract, and we can take your risk team through them before you sign.
Does “checks first” mean you avoid AI?
No. AI is used where judgment is needed, such as reading a messy page or choosing the closest match from a shortlist. It is not used for anything a rule, a lookup or arithmetic can settle, and an AI’s confidence score never approves anything on its own.
Does our data leave Bangladesh?
Not unless you allow it. Agents, databases, company memory, files and backups are hosted in Dhaka. If your policy allows an outside AI provider such as Claude or GPT for a task, the question and the context needed to answer it are sent to that provider, which does not run in Bangladesh.
Can we run it on our own servers?
Yes. It can run in a Tier-III data centre in Dhaka, on your own servers or private cloud, or on a hyperscaler. You decide per workload and can change it later.
Can we change AI provider later?
The platform is designed for that. Your company memory is kept apart from the AI provider, so moving to a new one is a planned change, not a rebuild of what your agents know.
Which parts are live, and which are early?
Agents, company memory and hosting in Dhaka are in production for clients today. The AI workbench is in early access.
Keep exploring
Data foundationOne correct record, documents read and checked.AI SubscriptionsLeading AI models, billed in Bangladesh.Agentic OperationsAgents that check, route and reconcile.

Bring us one process.
We’ll show you it running.

A 30-minute call with a Codemen architect. Bring one business problem. We map the workflow, show where rules decide and where a person approves, and what it costs to run in Bangladesh.

Book a 30-minute call Talk to an architect

Free. No slides unless you ask for them.