AITJE Coder
Describe a task in plain language. An agent works in its own copy of your repository, runs tests and submits the results for review. On a BOSGAME, Mac mini, local server, cloud server or external endpoint: you choose the environment, AITJE sets it up.
This applies to local features. If you use external models or online tools, data sharing, costs and dependencies may vary.
From €799 with basic installation, excl. VAT · View the price breakdown
Introductory offer for Assistant and Coder for the first 100 paying customers. Availability will be confirmed in your quotation. Hardware and server hire are charged separately.
Local AI · On edge · On premise
AITJE Coder is an AI Station with an autonomous coding agent that writes code, runs tests and prepares pull requests in your repositories. Local AI models run on edge or on premise on your own hardware or server: your code stays under your control, with no external API token costs for local model processing.
Coding agents, local models and management on your own computer or server.
Local AI · on-premises · managed in-house
AITJE Coder is a comprehensive environment for coding agents on your chosen hardware or server. The language model performs calculations within that environment or via a deliberately chosen external endpoint. Developers specify tasks in plain language, work with their own tools and approve each result before any changes are made to the repository.
“Correct the VAT rounding and ensure that all tests pass.”
You specify a task in plain language. The agent works independently in its own copy of the repository and submits the result for review. Suitable for tests, bugs, minor features and dependency updates.
Three ways of working on the same device. You can use them side by side.One input field, the repository and start branch next to it, and sample tasks below. On the left is what is ready for review.
Real product screenshots, featuring a fictional demo project. Choose the desktop or mobile view and explore both layouts.
The agent works in its own copy. A pull request is only sent to the repository once you have approved it.
Local models run on the box. The agent may iterate until the tests pass, without a counter at a cloud provider.
The web app works on any screen within your network. View a result and approve it from your mobile phone.
The interface is available in Dutch, English, German and French, in dark or light mode. You can issue commands in your own language.
What you ask for. What the code does.
A coding agent writes files, executes code and continues to work on the result. Take a look at a 3D scene, a small web app and a real bug fix. The task and the code are included.
“Create a premium 3D product presentation of a kinetic object: three brass rings around a glass core, on a dark plinth. Add realistic lighting, soft shadows, material choices, rotation and an exploded view. Make it responsive and ensure minimal movement.”
Three.js · JavaScript · WebGL
import * as THREE from "three";
import { OrbitControls } from "three/addons/controls/OrbitControls.js";
import { RoomEnvironment } from "three/addons/environments/RoomEnvironment.js";
// A complete procedural product scene. Every visible part is real geometry.
const stage = document.querySelector("#stage");
const scene = new THREE.Scene();
scene.background = new THREE.Color("#e8e9df");
const camera = new THREE.PerspectiveCamera(35, 1, 0.1, 60);
camera.position.set(4.3, 2.9, 5.8);
const renderer = new THREE.WebGLRenderer({ antialias: true, alpha: false });
renderer.setPixelRatio(Math.min(devicePixelRatio, 1.7));
renderer.shadowMap.enabled = true;
renderer.shadowMap.type = THREE.PCFSoftShadowMap;
renderer.toneMapping = THREE.ACESFilmicToneMapping;
renderer.toneMappingExposure = 1.35;
renderer.domElement.setAttribute(
"aria-label",
"Draaibaar kinetisch object met drie metalen ringen",
);
renderer.domElement.dataset.ready = "true";
renderer.domElement.tabIndex = 0;
stage.append(renderer.domElement);
const pmrem = new THREE.PMREMGenerator(renderer);
const environment = new RoomEnvironment();
const environmentMap = pmrem.fromScene(environment, 0.04);
scene.environment = environmentMap.texture;
environment.dispose();
pmrem.dispose();
const controls = new OrbitControls(camera, renderer.domElement);
controls.target.set(0, 1.65, 0);
controls.enablePan = false;
controls.enableZoom = false;
controls.enableDamping = true;
controls.minPolarAngle = Math.PI * 0.14;
controls.maxPolarAngle = Math.PI * 0.59;
const gold = new THREE.MeshPhysicalMaterial({
color: "#b99043",
metalness: 1,
roughness: 0.24,
clearcoat: 0.4,
});
const graphite = new THREE.MeshPhysicalMaterial({
color: "#242d28",
metalness: 0.72,
roughness: 0.31,
clearcoat: 0.35,
});
const glass = new THREE.MeshPhysicalMaterial({
color: "#b4d1c2",
metalness: 0.05,
roughness: 0.04,
transmission: 0.97,
thickness: 0.5,
ior: 1.45,
});
const warm = new THREE.MeshStandardMaterial({
color: "#e2ad47",
emissive: "#d89b25",
emissiveIntensity: 0.2,
metalness: 0.6,
roughness: 0.25,
});
const model = new THREE.Group();
scene.add(model);
renderer.domElement.addEventListener("keydown", (event) => {
if (!["ArrowLeft", "ArrowRight", "ArrowUp", "ArrowDown"].includes(event.key))
return;
event.preventDefault();
if (event.key === "ArrowLeft") model.rotation.y -= 0.15;
if (event.key === "ArrowRight") model.rotation.y += 0.15;
if (event.key === "ArrowUp")
camera.position.y = Math.min(camera.position.y + 0.2, 5);
if (event.key === "ArrowDown")
camera.position.y = Math.max(camera.position.y - 0.2, 1.5);
});
function part(geometry, material, parent = model) {
const mesh = new THREE.Mesh(geometry, material);
mesh.castShadow = true;
mesh.receiveShadow = true;
parent.add(mesh);
return mesh;
}
// Stepped pedestal, accent bands and a slender central support.
part(new THREE.CylinderGeometry(0.82, 0.9, 0.17, 96), graphite).position.y =
0.1;
part(new THREE.CylinderGeometry(0.815, 0.815, 0.018, 96), gold).position.y =
0.195;
part(new THREE.CylinderGeometry(0.69, 0.78, 0.11, 96), graphite).position.y =
0.25;
part(new THREE.CylinderGeometry(0.105, 0.19, 0.48, 48), gold).position.y = 0.52;
const assembly = new THREE.Group();
assembly.position.y = 1.86;
model.add(assembly);
const rings = [];
for (let index = 0; index < 3; index++) {
const pivot = new THREE.Group();
assembly.add(pivot);
const radius = 1.19 - index * 0.22;
const ring = part(
new THREE.TorusGeometry(radius, 0.045, 16, 144),
gold,
pivot,
);
ring.rotation.x = index === 1 ? Math.PI / 2 : 0;
// Visible precision marks around the outer gimbal.
for (let mark = 0; mark < 48; mark++) {
const angle = (mark / 48) * Math.PI * 2;
const tick = part(
new THREE.BoxGeometry(0.009, mark % 4 === 0 ? 0.048 : 0.022, 0.007),
graphite,
ring,
);
tick.position.set(
Math.sin(angle) * radius,
Math.cos(angle) * radius,
0.045,
);
tick.rotation.z = -angle;
}
for (const direction of [-1, 1]) {
const pin = part(
new THREE.CylinderGeometry(0.078, 0.078, 0.105, 32),
graphite,
pivot,
);
pin.rotation.z = Math.PI / 2;
pin.position.x = radius * direction;
part(new THREE.SphereGeometry(0.046, 20, 16), gold, pivot).position.x =
(radius + 0.065) * direction;
}
rings.push(pivot);
}
part(new THREE.IcosahedronGeometry(0.43, 4), glass, assembly);
part(new THREE.IcosahedronGeometry(0.21, 1), warm, assembly);
const meridian = part(
new THREE.TorusGeometry(0.46, 0.012, 12, 96),
gold,
assembly,
);
meridian.rotation.x = Math.PI / 2;
// A restrained floor and real soft contact shadows anchor the object.
const ground = part(
new THREE.PlaneGeometry(200, 200),
new THREE.ShadowMaterial({ color: "#526047", opacity: 0.17 }),
scene,
);
ground.rotation.x = -Math.PI / 2;
ground.position.y = -0.005;
ground.castShadow = false;
scene.add(new THREE.HemisphereLight("#ffffff", "#777f66", 2.4));
const key = new THREE.DirectionalLight("#fff5dd", 4.5);
key.position.set(-3, 7, 5);
key.castShadow = true;
key.shadow.mapSize.set(1536, 1536);
key.shadow.camera.left = -4;
key.shadow.camera.right = 4;
key.shadow.camera.top = 5;
key.shadow.camera.bottom = -3;
key.shadow.normalBias = 0.025;
key.shadow.blurSamples = 12;
scene.add(key);
const rim = new THREE.DirectionalLight("#e4edf6", 2.5);
rim.position.set(4, 3, -4);
scene.add(rim);
const motion = matchMedia("(prefers-reduced-motion: reduce)");
let playing = !motion.matches;
let exploded = 0;
let visible = true;
let lastTime = 0;
let elapsed = 0;
const toggle = document.querySelector("#motion");
function updateToggle() {
toggle.textContent = playing ? "Pauzeren" : "Draaien";
toggle.setAttribute("aria-pressed", String(playing));
}
toggle.addEventListener("click", () => {
playing = !playing;
updateToggle();
});
motion.addEventListener("change", () => {
playing = !motion.matches;
updateToggle();
});
updateToggle();
document.querySelector("#explode").addEventListener("input", (event) => {
exploded = Number(event.target.value) / 100;
document.querySelector("#explode-value").textContent =
`${event.target.value}%`;
});
for (const button of document.querySelectorAll("[data-finish]")) {
button.addEventListener("click", () => {
gold.color.set(button.dataset.finish);
document
.querySelectorAll("[data-finish]")
.forEach((item) =>
item.setAttribute("aria-pressed", String(item === button)),
);
});
}
document.querySelector("#reset").addEventListener("click", () => {
model.rotation.y = 0;
camera.position.set(4.3, 2.9, 5.8);
controls.target.set(0, 1.65, 0);
});
const resize = new ResizeObserver(() => {
const { width, height } = stage.getBoundingClientRect();
if (!width || !height) return;
renderer.setSize(width, height, false);
camera.aspect = width / height;
camera.updateProjectionMatrix();
});
resize.observe(stage);
new IntersectionObserver(([entry]) => {
visible = entry.isIntersecting;
}).observe(stage);
document.addEventListener("visibilitychange", () => {
lastTime = 0;
});
function frame(time) {
const delta = lastTime ? Math.min((time - lastTime) / 1000, 0.05) : 0;
lastTime = time;
if (visible && !document.hidden) {
if (playing) elapsed += delta;
rings.forEach((ring, index) => {
ring.rotation.set(
0.22 + Math.sin(elapsed * 0.32 + index * 1.4) * 0.52,
elapsed * (0.1 + index * 0.075) + index * 0.65,
index * 0.18,
);
ring.position.x = exploded * (index - 1) * 0.7;
ring.position.y = exploded * index * 0.12;
});
controls.update();
renderer.render(scene, camera);
}
requestAnimationFrame(frame);
}
requestAnimationFrame(frame);
Rotate the object, switch between brass, steel and copper, and see how the rings fit together. The source code shown here builds the scene you see.
The 3D scene and web app have been developed specifically for this page. They demonstrate types of programming work, not measured output from a specific local model run. The bug fix and product shots are taken from the actual Coder demo using fictitious company data. The model, task and controls determine the result.
From assignment to pull request
A task for AITJE Coder always follows the same five steps. The agent carries out the work in its own copy. You assess the result and decide what goes into the repository.
Select the repository and start branch, type the command and press Enter. Under Options, select the model, web search, links and the number of rounds. Use Schedule to set the same command for a specific time.
Actual screenshots from AITJE Coder, taken on 18 October 2026 on an AITJE box running the tiel-coder-35b model. The Bakery Demo and the invoicing project are fictional. The steps show three tasks from the same demo project; steps 3 to 5 belong to the same task.
Type a command, press enter and watch the agent read it. Viewed files are highlighted in yellow.
The status, the test report and the amended rules. Then the button that opens the pull request.
Aligning with your development process
A single code block is just the start. Coder links the task to the repository, the start branch, the modified files and the review. This allows your team to treat the result as a normal code change.
The agent works on the code you specify. In the diff, you can see every line that has been changed. The report details what has been examined, modified and tested. You can provide feedback on each file and request a retry.
The approved change will be recorded in your Git history. Colleagues can review it, existing CI can check the branch, and your own merge policy remains the final step. An open pull request does not automatically result in a merge or publication.
The administrator determines which repositories a developer is permitted to use and whether they may connect to repositories themselves. The agent works in their own working directory. To access Git, the agreed connection is set up, using tokens or a supported GitHub login.
Code, tests and history remain standard project files and Git changes. You can further amend the change, reject it or revert it via your own process. You do not need to copy the work from a chat to turn it into a reviewable change.
A coding agent with a steady workflow
Schedule maintenance, checks or a specific programming task on the server environment. The server will queue the job at the appropriate time; you can close your browser. The result will reappear amongst the tasks you can review.
Actual product recording using a fictitious demo project. The scheduled task is visible under ‘Scheduled’.
The server environment and task processing must be available. A schedule utilises the owner’s permissions and limits. External repositories and webhooks must be accessible. With multiple rounds, the agent can continue working on the task; assessment and publication remain part of the agreed review process.
Read. Build. Test. Improve.
A coding agent uses context, writes code and tries again. With a local model, this takes place on your own computing power – even if a task requires many model runs.
external token costs
per project and per million tokens
The agent does, however, use tokens. With local inference, you do not need to purchase tokens from an external AI provider. The available hardware, context and configured usage limits determine how much work can be processed at any one time.
Software, hardware, electricity and maintenance incur separate costs. If you deliberately choose an external model, the rates of that provider and the agreed budgets will apply.
View the full price breakdownQueue management, scheduling and rounds
Queue tasks or schedule them in advance. The agent will process them one by one. The result will be ready for review when you return.
One environment · desktop and mobile
Submit a task, view the result or review the modified files. You can do this on your phone too, via the same web app. The model and code remain on the chosen server environment; your phone is simply the point of access.
Within our own network. We set up external access via a VPN or agreed secure access and HTTPS.
Your own tools
The box speaks the language of the familiar tools. You only need to change the address and the key; your workflow remains the same.
A configuration is available for each of these tools. Long sessions in Claude Code, OpenCode, Codex and Aider are automatically summarised before they become unresponsive.
# eenmalig: de CLI van de box
curl -fsSL http://aitje-coder.local/connect/cli | bash
aitje-coder connect aitje-coder.local
# configuratie voor je tool, en een sessie met Git-context
aitje-coder config opencode
aitje-coder session --taak "Bug oplossen" opencode Prefer to click? The page /connect The box contains the certificate, your personal key and a ready-to-use configuration for each tool.
‘My sessions’ shows the work using your own tools, with model requests, tokens, repository and branch. Prompts and responses are not stored there.
Even when using the box as the address, some tools automatically contact their creator: a model catalogue, telemetry or a version check. This is neither a prompt nor code. We have measured this for each tool. /connect For each tool, it specifies the setting that disables it. We could not find a toggle for ‘Continue’.
Your environment, your code
AITJE Coder runs on a BOSGAME, a suitable Mac mini, your own local server, a cloud server or an external environment. We can install the application and models together or set up inference via a separate endpoint. You choose whichever option gives you greater independence; AITJE will tailor the hardware, access and management accordingly.
One agent task at a time. Comes as standard with tiel-coder-35b, which can store up to 128k tokens per conversation. Tested on actual hardware.
More memory for larger models and more concurrent usage. Performance measurements for this version will follow; we will fine-tune the capacity during the trial.
With a local model, tasks and code remain on the box. Programming with the installed models also works without an internet connection, provided the box and network are available.
AITJE Coder AI Station
A configured development environment for coding agents. The local model, the agent and the management system run on the agreed computer or server. Developers connect their browser, terminal or editor to that environment.
In the standard appliance installation, PostgreSQL runs as a local service within the Coder environment, with its own permanent data volume. This contains users, tasks, schedules, review data and usage statistics. Repositories and the agent’s working folders are stored separately on local storage.
The browser runs the web app. A terminal or editor can use the model on the box with a personal key; that client’s files may be stored on the developer’s workstation. When performing tasks via the Coder web app, the agent works in their own working directory on the box.
Who is authorised to submit or approve tasks, which repositories a developer is permitted to use, which models and tools are accessible, and what limits apply. Budgets, private keys, support access and the log are managed centrally.
Computing power and agent tasks managed in-house, using the team’s repositories and review procedures. AITJE selects and tests models and configures the tools, permissions and queue. Your team can build without needing an external token account for local inference.
The local model is available without an external model API. Working with locally hosted code can continue as long as the box, local services and network are functioning. However, a new clone, package download or pull request to an external Git service may have to wait for that service to become available. Local inference and external Git accessibility are separate dependencies.
postgres-data-volume. /var/aitje/repos en /var/aitje/work. AI Station describes the configured combination of environment, model and software. The software price, installation and required hardware or hosting are listed separately in the price breakdown and quotation.
Choose your computing power and your model provider
There are two options for where AITJE runs and where the language model performs its calculations. Whether it’s all on your own hardware, a model on your server or a linked inference endpoint: we’ll set up the combination that best suits your work.
Access, working method and usage
AITJE Coder is more than just a model on a server. The management environment determines who is permitted to work on which code, using which methods and tools, and within what limits.
Connect a repository once and manage access on a per-repository basis: log in to GitHub using a code, a token, or revoke access. Without access, the agent can read a public repository but cannot open a pull request.
Every developer is given their own account and a personal key with their own limits. Invitations are sent via a link or QR code that is valid for 24 hours. Every administrative action is recorded in plain language in the log.
In a single edit
Repositories, skills, links, budgets and sessions. The yellow border always indicates the rule that determines what the agent is permitted to do.
From the first task to day-to-day management
The agent is just the start. Repositories, models, permissions, scheduling, skills, integrations and management make it useful for your team. See what Coder has to offer.
New task · My sessions · Queue & schedule · Settings
Select the repository, start branch, model and web search. Where necessary, work on linked repositories within your own secure working folder.
Track files, commands, tests and errors. Stop a task with confirmation; a stalled loop will be given a limited number of restarts.
View changes side by side or one below the other. Provide an assessment and a comment for each file. Compare runs or have a change re-run.
A reviewer opens the pull request on GitHub, GitLab or Gitea. GitHub login using a code and a fork where write access is lacking are supported.
Queue or pause the process. Schedule it once, daily, weekly or via cron. Start on a push or a GitHub/Linear issue and run multiple rounds.
If you are working in an editor or chat yourself, this takes precedence over batch tasks. The queue shows what is pending, running or awaiting review.
Actual log: report and modified files from the local bug fix.
Based on current product features. We will coordinate the availability of models, external services, hardware capacity and additional setup. The screenshots shown contain fictitious company data; the agent tasks in those screenshots were actually carried out.
How well do local models programme already?
The local agent has already fixed bugs, added features and written tests. You can see this in the actual product footage. When selecting a model, we also look at our own tests and published benchmarks, taking into account the task and hardware involved.
Three Python tasks, each run twice per model, using the actual Coder stack. Hidden tests checked discount codes, a rounding bug and a CLI tool. This allows you to assess a completed task rather than just the speed of individual tokens.
BOSGAME M6 · 32 GB. Small trial on a single repository: six runs per model; no overall success rate for all programming tasks. Turnaround times include the model change and are indicative.
Median time from submission to final status: 553 seconds
Median time from submission to final status: 384 seconds
Median time from submission to final status: 831 seconds
These figures from Qwen’s model map highlight various strengths. They measure a different configuration to the AITJE test above. A score does not automatically predict the result for your code or for a specific BOSGAME.
Scale 0–100. Qwen’s own evaluation; measurement set-ups and agent configurations vary. For SWE-bench Pro, the Opus score has been taken from the official report. This is not a BOSGAME measurement and does not guarantee that Qwen 3.8 is the standard on every Coder configuration. Quantisation, context, tools and memory influence real-world performance.
Source: Qwen3.8-27B model card · verified 18 October 2026Coder keeps the choice of model separate from the agent, the repository link and the review. A new or smaller model may be worth considering for speed, memory or quality. AITJE tests suitable models with similar tasks before recommending a new configuration. The benefits and the timing of these are not guaranteed in advance; updates and model configuration are agreed separately.
From the initial consultation to the first pull request
We select and test the configuration based on your tasks. We assess quality, waiting times and concurrent usage using representative workloads.
We look at which repositories are involved, who is working with them and which tasks you wish to outsource. This enables us to select the box and model that best suit your needs.
Book a demo with your codeAITJE servers · inference in the data centre
Don’t need your own hardware? AITJE offers multiple servers in data centres running language models and coding agents. Use a pre-configured environment or perform inference via an endpoint, specifically tailored to your workflow.
Discuss an AITJE serverCost-saving target for suitable workflows, compared with the use of frontier APIs from OpenAI, Anthropic and Gemini. We assess the model quality, volume and costs against your specific tasks; the rationale and rates will be included in your proposal.
Hosting, computing power and inference are quoted separately. The achievable savings vary per workflow and do not constitute a general price guarantee.Investing in your own coding agents
You purchase AITJE Coder per environment. For local model use, there is no external token account. However, hardware, energy and maintenance are included in the total costs; we make these clear in advance.
Ongoing use of the delivered version. The number of concurrent users and tasks falls within the capacity of the chosen environment.
€799
Product + basic installation · excl. VAT
Hardware or hosting is charged separately · subject to quotation
Standard from €898. Introductory offer for Assistant and Coder for the first 100 paying customers. Availability will be confirmed in your quotation.
Let AITJE take care of itInstallation on a pre-configured, suitable hardware or server environment, testing the system and providing guidance. We agree on a brief follow-up service. The basic installation is estimated to take two to two and a half hours.
You can choose your hardware and hosting separately
The product price shown above is for software and basic installation. New hardware, server hire and paid inference are charged separately. If you have existing suitable hardware, there may be no need to purchase a new device.
BOSGAME M5 / M6 · Mac mini · own hardware · own server · Hetzner · DigitalOcean · AITJE data centre
Whether it’s your own hardware, an external server, the cloud or a combination: tell us what you want to become more independent from. AITJE will find a suitable technical and financial setup.
Solution and total investment to be quoted. All amounts are exclusive of VAT; external providers and ongoing use are agreed separately.
Request this configurationLocal models, the agent with reviews and pull requests, queuing and scheduling, integration with your own tools, and management.
There are no external token costs for local models. If you allow external models, the provider will charge for them directly; budgets limit this.
Support and maintenance are optional, starting from €49 per month. Updates and new models are agreed separately.
Case studies
From searching through internal knowledge to connecting work processes. Discover the application, the need and the approach.
Development agency / Assistant, Coder & model management
AITJE set up a shared AI environment for a development agency, featuring Assistent, Coder and ten models to choose from. A shared token pool of €350 per month replaces nine subscriptions to €200. Model usage and token costs per task can be tracked via Asana.
View the approach
Marketing agency / level design & in-house coding agency
A marketing agency took over an existing mobile puzzle game but ran into difficulties when creating new levels. AITJE analysed the game, built a reusable level design system and linked AITJE Coder to Qwen on its own server. Now, the agency 25 creates new levels every week from a single assignment in OpenCode.
View the approach
Power leads / 3D catalogue & workflow optimisation
AITJE built a workflow for a partner specialising in contact cables that converts product data and technical drawings from WordPress into 3D models. AITJE Coder controls Blender; Three.js makes the GLB files usable on the website. Following optimisation, API costs fell from approximately €9 to €0.11–€0.18 per product.
View the approachTechnical details
Working with the installed models, programming via GUI, API, CLI and locally available frameworks, and reading, modifying and executing local projects.
Searching for online documentation, downloading new models, using packages and Git services, and remote support where agreed.
Speed, model variant, context capacity and the number of concurrent users depend on the hardware. A larger team may require a server or more than one environment.
Configurations are available for Claude Code, OpenCode, Codex CLI, Aider, Cline, Roo Code, Kilo Code, Continue and Zed. Claude Code communicates directly with the box’s Anthropic Messages API, whilst the other tools use an OpenAI-compatible endpoint. Coder does not guarantee compatibility with every tool.
Frequently Asked Questions
A dedicated environment for coding agents: a local alternative to services such as Claude Code and Codex. Agents read and modify your code, execute commands and complete tasks iteratively.
Yes. Claude Code, OpenCode, Codex, Aider, Cline, Continue and other tools connect to your own model environment. The /connect page in the AITJE environment provides the certificate, your private key and a ready-to-use configuration for each tool. You assign tasks to the agent via the web app.
The standalone product costs €699 per environment, with no per-user licence fee. Including basic installation, testing, training and agreed short-term aftercare, an introductory price of €799 applies; the standard price starts at €898. Introductory offer for Assistent and Coder for the first 100 paying customers. Availability will be confirmed in your quotation. Hardware, server hire and additional setup are quoted separately. All amounts are exclusive of VAT.
For many tasks, certainly – especially if you keep them simple and clear. See how a marketing agency uses AITJE Coder 25 to create new game levels every week, or request a demo to discuss the possibilities for your own work.
Yes. Adding or removing a local model is done within your model environment; AITJE can do this for you either on a stand-alone basis or via an SLA. An administrator can also add an external provider. By default, these models are not accessible to anyone: you grant access to them on a per-user basis.
You can work with all programming languages and frameworks; Coder is not limited to a single fixed stack. You can provide additional documentation, instructions and project agreements via ‘skills’. The quality depends on the chosen model and the available context, so for a specific stack, we will test it against your tasks.
Yes. Coder can read, interpret and adapt existing projects. So you don’t need to start a new project. By providing project documentation and clear instructions, you give the agent context regarding the structure and conventions within your codebase.
For tasks carried out via the web app, the agent operates in a secure container with its own copy of the repository. It is permitted to install packages, but cannot access your internal network. You determine access on a per-repository basis; for GitHub, you log in once using a code, without having to create a token yourself. If you use your own tool, such as Claude Code or OpenCode, that tool determines which files and commands the agent is permitted to use.
Yes. For tasks carried out via the web app, nothing changes until you approve the result. You read the agent’s report and see every line that has been changed. Approving opens a pull request on GitHub, GitLab or Gitea; ‘Changes required’ rejects the request or has the agent start again. When using the tool interactively, your own tool determines when approval is required.
Yes. Coder can write tests, run existing tests and investigate errors. To do this, the agent can use your code, error messages and test results to suggest a change. You can then assess the tests and changes yourself.
No. The environment can run on a BOSGAME M5/M6, a suitable Mac mini, existing hardware, a local server, a cloud server with, for example, Hetzner or DigitalOcean, or an AITJE server. The application, agents and model processing can be set up together or separately. AITJE will find the environment that suits your workflow and desired level of independence.
Yes. You can schedule tasks as one-off or recurring, with supported timetables and push or issue events. The server queues the work. The agent works in their own working copy, displays changes and checks, and prepares a pull request after review. The required server, permissions, models and Git services must be available.
The product price quoted, including basic installation, is not a total price inclusive of hardware. New hardware, server hire, additional setup, support and any external inference are quoted separately. If you choose your own model running on your own computing power, there is no external API token charge; for paid providers, their rates apply.
An administrator can add a supported external model or OpenAI-compatible endpoint, including a suitable Nebius Token Factory integration. AITJE also offers its own data centre servers for models, coding agents and workflow-oriented inference. We test model quality and tool usage against your tasks and set out access, capacity and costs in advance.
Yes. Request a visit with hardware or choose a demo endpoint for a remote trial. The contact form will automatically include Coder and your chosen demo option. You can also specify your preference for hardware or hosting there.
Wider possibilities
AITJE Coder See what AITJE can build. Need different workflows, integrations or hardware? That’s possible with AITJE Custom — bespoke AI.
Visit AITJE CustomMOVING FORWARD TOGETHER WITH AI
Let AITJE select the hardware and deliver everything ready for use, or see it for yourself first in a personalised demo.
