OpenEng Labs

OpenEng Labs · Bangalore, India

AI automation for enterprises, private by construction.

OpenEng Labs is an AI research company in Bangalore that builds agentic software on open-source foundations and brings AI automation to enterprises.

Two product lines carry that work. OpenEng Engineering is five free developer apps in beta at openeng.app (opens in a new tab), plus a separate AI engine for open-weight models; both run on your own machine. AEL, at ael.openeng.ai (opens in a new tab), is a compiled language built toward agent engineering; its agent features are planned. A services team adapts open-source AI tooling to the way your business already runs.

Talk to usSee the products

Product status

Status, as of September 25, 2026

  • Beta

    Engineering apps

    Five apps in beta, free: Terminal, Kubernetes, Data, Git, Chat

Status, as of September 25, 2026

  • Live

    AI engine

    Live on Linux x86_64; macOS and Windows builds coming; GA planned Q4 2026

  • Planned

    AEL 0.1

    In development, not released; linux-x86_64 is the sole qualified target

What we do

Research, products, and services, from one team.

Each line of work feeds the next: research programs are documented against running code, the products are their applied end, and the services team puts both to work inside your systems.

  • 5 programs

    Research

    Applied research in agentic systems, from governed execution to agent languages. Each program is written up against the running code, including what is not built yet.

    Research programs

  • 2 product lines

    Products

    OpenEng Engineering: five free developer apps and a separate AI engine for open-weight models. AEL: a compiled language built toward agent engineering.

    OpenEng Engineering

  • 6 offers

    Services

    Six offers, from AI automation support to research collaboration. We start from the open-source tools you already run and verify the work on your machines.

    Services

Products

Two product lines, both on hardware you control.

OpenEng Engineering puts a thin browser surface over engines on your own machine. The AEL compiler runs offline and emits a native executable.

OpenEng Engineering

Developer apps and a separate AI engine, both local-first.

OpenEng is a suite of five professional developer apps — Terminal, Kubernetes, Data, Git, and Chat — that run in your web browser but execute on a small native engine you run on your own machine.

A separate AI engine downloads and runs open-weight models on your machine. It is live on Linux x86_64, and general availability is planned for Q4 2026.

  • Terminal
  • Kubernetes
  • Data
  • Git
  • Chat

Free, every feature · one account signs in to all five

terminal.openeng.app
$git status --short
 M src/engine.rs
?? notes.md
exit 038 ms~/work/engine

Explore OpenEng Engineering

AEL

ael.openeng.ai

A compiled language, built toward agent engineering.

AEL is a native language whose compiler is first-party Rust: source to binary AEL IR to a native executable, with no LLVM and no generated C. Version 0.1 is in development and qualifies linux-x86_64 alone.

The compiler has zero external dependencies, and its linux-x86_64 output is a static executable that needs only the kernel. The agent features, including the runtime and the package registry, are planned.

0.1: 22 of 48 acceptance boxes closed · status record of September 19, 2026

ael
$ ael --version
ael 0.1.0-dev
main.ael
main main() -> i64 {
    let total: i64 = 6 * 7;
    print(total);
    return total;
}
  • main.ael
  • .aelir
  • ELF

Visit ael.openeng.ai (opens in a new tab)

Why private by construction

Privacy that is built in, not promised.

Every OpenEng surface is designed so that your code, clusters, databases, models, and secrets stay on hardware you control. Each card is a property of the architecture, tagged with its status as of September 25, 2026.

  • Engines on your machine

    The AI engine downloads and runs open-weight models entirely on your machine. The five apps drive a small native engine on your own hardware. In both cases the browser is only the surface.

    apps beta · AI engine live

  • Sealed channels

    Browser-to-engine calls are sealed end to end, except the session handshake. At boot the engine checks releases.openeng.app for an update and makes a one-time sign-in key exchange; your code and data never reach an OpenEng server.

    five apps · beta

  • No telemetry in the AI engine

    The AI engine binds no inbound port and sends no telemetry. Models, threads, memory, and secrets stay on the machine that runs them.

    AI engine · live

  • Secrets by reference

    The Console never runs anything and never stores a secret. Credentials are environment-variable placeholders, a pasted raw secret is rejected, and each configuration is published as an immutable id@version.

    Console · live

  • Models without API keys

    The 28 models in the on-device catalog, from 1B to 70B parameters, run on your hardware with no API key, so prompts and outputs are never sent to a model provider.

    models catalog · live

  • Cloud tools in your own account

    Cells ship as source only, never as binaries. You deploy them into your own AWS, Azure, or GCP account, behind your cloud’s API gateway, and can read every line before they run.

    cells catalog · live

Built on and works with

llama.cpp · GGUFOllamaHugging Face HubONNX RuntimeMCPKubernetesPostgreSQLRedisNeo4jMongoDBApache KafkaRustReact

How the AI engine executes

Everything passes through one aperture.

Inside the OpenEng AI engine the model never executes anything itself. It emits typed intents, and every side effect runs behind one governed gateway with a fail-closed policy floor that no configuration can remove.

Inside the OpenEng AI engine the model emits typed intents; every side effect passes through one governed gateway with a fail-closed policy floor before it reaches the shell, connectors, cells, or MCP servers. The engine binds no inbound port and sends no telemetry.OpenEng AI engineno inbound port · no telemetryModelemits typed intentsGoverned gatewayfail-closed policy floorShellkernel sandbox by defaultConnectorsCellsAWS, Azure, GCPMCP servers

  • Typed intents only

    The model proposes; the governed engine disposes. Nothing reaches the shell, a connector, a cell, or an MCP server without passing the gateway.

  • Fail-closed floor

    A build-time gate enforces the no-bypass rule. Chosen plugins run above the floor, and a configuration can add policy but never remove it.

  • Kernel sandbox

    Shell commands run in an OS kernel sandbox by default wherever the machine provides one: a read-only root, writes only inside the workspace, and the network off unless you allow it.

This diagram describes the OpenEng AI engine. The five Engineering apps work differently: each drives a native engine on your machine over a sealed loopback channel.

Evidence

Numbers we can show you the source for.

Every figure below is read from a public catalog, status page, or product site, and each one links to its source.

5

Developer apps, free and in beta

Source: openeng.app

3,300+

Building blocks in six AI-platform catalogs

Source: agents.openeng.app and five more catalogs

1,503

AEL compiler tests passed, 0 failed

Source: ael.openeng.ai/status

Counts read from each catalog.json and product site on September 25, 2026; the AEL figure is from the status record of September 19, 2026. The AI platform is not launched: a Linux x86_64 engine is live, and general availability is planned for Q4 2026.

Services

Services that start from your open-source stack.

We begin with the open-source and open-weight tools you already run, and build what is missing.

See all services

  • AI Automation Support

    We design, run, and support recurring automations on the OpenEng AI engine with loops, cells, and connectors. Schedulers and headless runs will arrive with the paid plans in Q4 2026.

    Live

  • Custom AI Solutions

    Agents built for your business as authored configs, drawing on the agent, skill, and template catalogs and published as one immutable id@version.

    Live

  • Open-Source and Open-Weight Customization and Support

    We adapt and support the open-weight models and open-source tools you run, such as Ollama, the Hugging Face Hub, ONNX, and MCP servers.

    Live

  • Private and On-Prem Agent Platforms

    The local-first architecture is live today. Enterprise and on-prem plans will launch in Q4 2026.

    Planned for Q4 2026

  • Agent Engineering with AEL

    AEL is a compiled language built toward agent engineering, with a first-party Rust compiler. Version 0.1 is in development, and its agent features are planned.

    In development

  • Research Collaboration

    Partner with us on applied agentic-systems research: architecture surveys, code-verified gap analyses, sandboxing, and human-in-the-loop design.

    Live

How we work

Four steps from discovery to production.

Scope is agreed in writing, every run is verified on your machines before it goes live, and at every step we say plainly what is live today and what is planned.

  1. 01

    Discover

    We map the systems in play and the automation candidates: what repeats, what it touches, and who signs off.

  2. 02

    Design

    We choose the models, the connectors, and the governance policy, and decide what runs on your machines.

  3. 03

    Build

    We author the configs, agents, and customizations on your stack and verify every run on your machines.

  4. 04

    Operate

    We support the automations in production, keep the runbooks current, and iterate as the work changes.

Research

Five research programs behind the products.

The products are the applied end of five research programs, each documented against the running code, including what is not built yet.

Explore the research programs

  • Governed agentic execution

    A supervisor loop where the model only emits typed intents and every side effect passes one governed gateway.

    In the engine today

  • Open-weight model execution

    Local-first execution of downloadable open-weight models: GGUF, Ollama, ONNX, and network runtimes out of the box, and vLLM and six more once their toolchain is installed.

    Linux x86_64 engine live

  • Sandboxed execution

    Four cooperating layers, with kernel-enforced isolation for each shell command and the per-user daemon. The shell sandbox is on by default wherever the OS provides one.

    Shell sandbox live

  • Human-in-the-loop

    Durable suspend, human approval, and resume at the call boundary. Suspend exists in the gateway today; the approval flow will arrive with the paid plans in Q4 2026.

    Approval flow planned

  • Agent languages

    AEL, a compiled language built toward agent engineering. Version 0.1 of the compiler is in development; the agent features, including the runtime, are planned.

    AEL 0.1 in development

FAQ

Three questions we hear first.

Read the full FAQ

What is OpenEng Labs?

OpenEng Labs is an AI research company in Bangalore that builds agentic software on open-source foundations and brings AI automation to enterprises. Its two product lines are OpenEng Engineering, which is five developer apps at openeng.app plus a separate AI engine, and AEL at ael.openeng.ai. A services team adapts open-source AI tooling to enterprise needs.

Is OpenEng open source?

No. The OpenEng platform itself is not open source. Its engine runs open-weight (open-source) models such as Llama, Qwen, and Gemma entirely on your machine, and the company builds on and customizes open-source software such as Ollama, ONNX, and MCP for enterprises.

What is the difference between OpenEng Engineering and AEL?

OpenEng Engineering is available today: five free developer apps in beta at openeng.app, plus a separate AI engine for open-weight models that is live on Linux x86_64. The Terminal, Kubernetes, and Data engines can be installed now; Git and Chat engine builds are not published yet. AEL, at ael.openeng.ai, is a compiled, statically typed language with a first-party Rust compiler. Agent engineering is its stated goal, and its agent features are planned. Version 0.1 is in development, not released, and qualifies linux-x86_64 alone.

Tell us what you want automated.

Bring the workflow, the systems it touches, and the constraints it must respect. We will say plainly what is live today and what is planned.

Talk to usTry the apps at openeng.app (opens in a new tab)