


On-Prem — Nothing Leaves the Network
Clients: VS Code, IntelliJ, iCli
Audited, Read-Only Agent Tools
AI Workflow
iChat by IOTA
Our own product, built like bank software: an air-gapped AI engineering assistant — three clients, one governed platform, and nothing ever leaves your network.
Year
2026
Industry
AI for Regulated Finance
SERVICE USED
AI Development, AI Agent Development, Digital Engineering
Challenge
Cloud AI assistants are unusable where source code is a regulated asset
(qtf® — the problem)
Cloud AI assistants are transformative — and blocked wherever source code is a regulated asset. Banks needed the capability without the egress.

Play
Inside iChat
Platform overview
(qtf® — solution)
The problem: the best engineers in regulated finance were locked out of the AI decade.
Cloud AI assistants are transformative — and unusable where prompts carry proprietary code, credentials, and customer context. Banks told us the same three things: data egress is a compliance event, auditors need evidence, and procurement needs provenance. So we built the assistant we could not buy.
One gate, everything behind it.
iChat puts a single service tier between every client and the model. Identity (OIDC/SSO), role-based model policy, offline license checks, and a full audit trail are enforced server-side — before a single token reaches the model. The models themselves run on vLLM, on your GPUs, inside your perimeter. There is no telemetry and no phone-home: the platform verifies its own license cryptographically, offline.
An agent that investigates before it answers.
A LangGraph-orchestrated agent searches, reads, outlines, and maps your repositories through seven read-only, path-jailed tools — exact search, semantic search, repository maps, file reads and structure — with every call streamed live to the developer and written to the audit log. Retrieval is hybrid (keyword plus semantic, rank-fused) and scales from a single file store to SQLite to ClickHouse without changing the product.
iCli: the whole platform, in a terminal.
Backend teams live in the shell, so iChat’s third client is a terminal UI with the same enforcement as the IDEs: streaming chat with live tool activity, headless OIDC device-flow sign-in for SSH sessions, and a one-shot mode that pipes governed AI straight into scripts and CI. No model keys on the laptop, nothing to bypass.
Licensed for the real world.
Licenses are signed files verified locally — evaluation licenses renew monthly, production licenses renew yearly, and a grace period means an expiry never strands engineers mid-sprint. Seats, features, and tenants are all in the signed payload; renewal is dropping in a new file. The stack underneath is deliberately open and clean: LangGraph, vLLM, ripgrep, and tree-sitter, with a full third-party license manifest and zero copyleft.
iChat is available today for evaluation inside your own perimeter — talk to us and see it answer questions about your codebase, on your hardware.
(qtf® — Technology Stacks)
Cogni
Tenso
MindX
Pulse
GridX
NovaA
(qtf® — client review)
We built iChat the way we advise banks to build: air-gapped, audited end to end, and licensed to run with the network cable unplugged.

IOTA Labs Team
Data & AI Practice



On-Prem — Nothing Leaves the Network
Clients: VS Code, IntelliJ, iCli
Audited, Read-Only Agent Tools
AI Workflow
iChat by IOTA
Our own product, built like bank software: an air-gapped AI engineering assistant — three clients, one governed platform, and nothing ever leaves your network.
Year
2026
Industry
AI for Regulated Finance
SERVICE USED
AI Development, AI Agent Development, Digital Engineering
Challenge
Cloud AI assistants are unusable where source code is a regulated asset
(qtf® — the problem)
Cloud AI assistants are transformative — and blocked wherever source code is a regulated asset. Banks needed the capability without the egress.

Play
Inside iChat
Platform overview
(qtf® — solution)
The problem: the best engineers in regulated finance were locked out of the AI decade.
Cloud AI assistants are transformative — and unusable where prompts carry proprietary code, credentials, and customer context. Banks told us the same three things: data egress is a compliance event, auditors need evidence, and procurement needs provenance. So we built the assistant we could not buy.
One gate, everything behind it.
iChat puts a single service tier between every client and the model. Identity (OIDC/SSO), role-based model policy, offline license checks, and a full audit trail are enforced server-side — before a single token reaches the model. The models themselves run on vLLM, on your GPUs, inside your perimeter. There is no telemetry and no phone-home: the platform verifies its own license cryptographically, offline.
An agent that investigates before it answers.
A LangGraph-orchestrated agent searches, reads, outlines, and maps your repositories through seven read-only, path-jailed tools — exact search, semantic search, repository maps, file reads and structure — with every call streamed live to the developer and written to the audit log. Retrieval is hybrid (keyword plus semantic, rank-fused) and scales from a single file store to SQLite to ClickHouse without changing the product.
iCli: the whole platform, in a terminal.
Backend teams live in the shell, so iChat’s third client is a terminal UI with the same enforcement as the IDEs: streaming chat with live tool activity, headless OIDC device-flow sign-in for SSH sessions, and a one-shot mode that pipes governed AI straight into scripts and CI. No model keys on the laptop, nothing to bypass.
Licensed for the real world.
Licenses are signed files verified locally — evaluation licenses renew monthly, production licenses renew yearly, and a grace period means an expiry never strands engineers mid-sprint. Seats, features, and tenants are all in the signed payload; renewal is dropping in a new file. The stack underneath is deliberately open and clean: LangGraph, vLLM, ripgrep, and tree-sitter, with a full third-party license manifest and zero copyleft.
iChat is available today for evaluation inside your own perimeter — talk to us and see it answer questions about your codebase, on your hardware.
(qtf® — Technology Stacks)
Cogni
Tenso
MindX
Pulse
GridX
NovaA
(qtf® — client review)
We built iChat the way we advise banks to build: air-gapped, audited end to end, and licensed to run with the network cable unplugged.

IOTA Labs Team
Data & AI Practice



On-Prem — Nothing Leaves the Network
Clients: VS Code, IntelliJ, iCli
AI Workflow
iChat by IOTA
Our own product, built like bank software: an air-gapped AI engineering assistant — three clients, one governed platform, and nothing ever leaves your network.
Year
2026
Industry
AI for Regulated Finance
SERVICE USED
AI Development, AI Agent Development, Digital Engineering
Challenge
Cloud AI assistants are unusable where source code is a regulated asset
(qtf® — the problem)
Cloud AI assistants are transformative — and blocked wherever source code is a regulated asset. Banks needed the capability without the egress.

Play
Inside iChat
Platform overview
(qtf® — solution)
The problem: the best engineers in regulated finance were locked out of the AI decade.
Cloud AI assistants are transformative — and unusable where prompts carry proprietary code, credentials, and customer context. Banks told us the same three things: data egress is a compliance event, auditors need evidence, and procurement needs provenance. So we built the assistant we could not buy.
One gate, everything behind it.
iChat puts a single service tier between every client and the model. Identity (OIDC/SSO), role-based model policy, offline license checks, and a full audit trail are enforced server-side — before a single token reaches the model. The models themselves run on vLLM, on your GPUs, inside your perimeter. There is no telemetry and no phone-home: the platform verifies its own license cryptographically, offline.
An agent that investigates before it answers.
A LangGraph-orchestrated agent searches, reads, outlines, and maps your repositories through seven read-only, path-jailed tools — exact search, semantic search, repository maps, file reads and structure — with every call streamed live to the developer and written to the audit log. Retrieval is hybrid (keyword plus semantic, rank-fused) and scales from a single file store to SQLite to ClickHouse without changing the product.
iCli: the whole platform, in a terminal.
Backend teams live in the shell, so iChat’s third client is a terminal UI with the same enforcement as the IDEs: streaming chat with live tool activity, headless OIDC device-flow sign-in for SSH sessions, and a one-shot mode that pipes governed AI straight into scripts and CI. No model keys on the laptop, nothing to bypass.
Licensed for the real world.
Licenses are signed files verified locally — evaluation licenses renew monthly, production licenses renew yearly, and a grace period means an expiry never strands engineers mid-sprint. Seats, features, and tenants are all in the signed payload; renewal is dropping in a new file. The stack underneath is deliberately open and clean: LangGraph, vLLM, ripgrep, and tree-sitter, with a full third-party license manifest and zero copyleft.
iChat is available today for evaluation inside your own perimeter — talk to us and see it answer questions about your codebase, on your hardware.
(qtf® — Technology Stacks)
Cogni
Tenso
MindX
Pulse
GridX
NovaA
(qtf® — client review)
We built iChat the way we advise banks to build: air-gapped, audited end to end, and licensed to run with the network cable unplugged.

IOTA Labs Team
Data & AI Practice



On-Prem — Nothing Leaves the Network
Clients: VS Code, IntelliJ, iCli
Audited, Read-Only Agent Tools
AI Workflow
iChat by IOTA
Our own product, built like bank software: an air-gapped AI engineering assistant — three clients, one governed platform, and nothing ever leaves your network.
Year
2026
Industry
AI for Regulated Finance
SERVICE USED
AI Development, AI Agent Development, Digital Engineering
Challenge
Cloud AI assistants are unusable where source code is a regulated asset
(qtf® — the problem)
Cloud AI assistants are transformative — and blocked wherever source code is a regulated asset. Banks needed the capability without the egress.

Play
Inside iChat
Platform overview
(qtf® — solution)
The problem: the best engineers in regulated finance were locked out of the AI decade.
Cloud AI assistants are transformative — and unusable where prompts carry proprietary code, credentials, and customer context. Banks told us the same three things: data egress is a compliance event, auditors need evidence, and procurement needs provenance. So we built the assistant we could not buy.
One gate, everything behind it.
iChat puts a single service tier between every client and the model. Identity (OIDC/SSO), role-based model policy, offline license checks, and a full audit trail are enforced server-side — before a single token reaches the model. The models themselves run on vLLM, on your GPUs, inside your perimeter. There is no telemetry and no phone-home: the platform verifies its own license cryptographically, offline.
An agent that investigates before it answers.
A LangGraph-orchestrated agent searches, reads, outlines, and maps your repositories through seven read-only, path-jailed tools — exact search, semantic search, repository maps, file reads and structure — with every call streamed live to the developer and written to the audit log. Retrieval is hybrid (keyword plus semantic, rank-fused) and scales from a single file store to SQLite to ClickHouse without changing the product.
iCli: the whole platform, in a terminal.
Backend teams live in the shell, so iChat’s third client is a terminal UI with the same enforcement as the IDEs: streaming chat with live tool activity, headless OIDC device-flow sign-in for SSH sessions, and a one-shot mode that pipes governed AI straight into scripts and CI. No model keys on the laptop, nothing to bypass.
Licensed for the real world.
Licenses are signed files verified locally — evaluation licenses renew monthly, production licenses renew yearly, and a grace period means an expiry never strands engineers mid-sprint. Seats, features, and tenants are all in the signed payload; renewal is dropping in a new file. The stack underneath is deliberately open and clean: LangGraph, vLLM, ripgrep, and tree-sitter, with a full third-party license manifest and zero copyleft.
iChat is available today for evaluation inside your own perimeter — talk to us and see it answer questions about your codebase, on your hardware.
(qtf® — Technology Stacks)
Cogni
Tenso
MindX
Pulse
GridX
NovaA
(qtf® — client review)
We built iChat the way we advise banks to build: air-gapped, audited end to end, and licensed to run with the network cable unplugged.

IOTA Labs Team
Data & AI Practice
(iota — 05)
More cases



(iota — 15)
OUR PRINCIPLES
Book a
free call
We design AI systems that improve real work —
not just demonstrate technology.
We’ll review your
workflows, identify
where AI can create
impact, and outline
a clear path forward.
We’ll review your workflows, identify AI opportunities, and outline a clear path forward.
We’ll review your workflows, identify where AI can create impact, and outline
a clear path forward.
No preparation needed — we’ll guide the conversation
and focus on what matters.
No preparation needed — we’ll guide the conversation and focus on what matters.




4 AI products
100%
on-prem, air-gapped AI
1.5k reviews


