Aerenium. On your computer. On your side.
Aerenium is a cross-platform desktop app that carries multi-step work forward with current-generation models. Your workspace data stays on your machine, and you control permissions, pace, and spend.
Aerenium is in alpha (0.5). Free to try; things will break; your feedback shapes 1.0. Share feedback with the community.
Ask once. Keep the final say.
Set budget ceilings, pause instantly, or stop the whole run from the breaker board. Every step stays inspectable.
Instead of saving everything equally, it learns which combinations of past context help produce useful results. Helpful combinations return; noisy ones fade. You can inspect and manage every layer.
Learns useful combinations. Consolidates over time. Stays inspectable.
Agent Mode carries a multi-step job forward inside your limits. A sandboxed swarm can split larger work while the main conversation stays focused.
Budget ceilings. Sticky pause. Breaker board. Sandboxed specialists.
Choose permissions and approvals, cap model spend, pause instantly, or stop every worker from one breaker board.
Permission gates. Spend ceilings. One place to stop everything.
Most AI tools stop at the answer.
The work loop
Start with the outcome. Aerenium can bring in relevant files and context, branch into research, hand parts to specialist workers, and keep you close to the decisions and controls that matter.
Describe the outcome in ordinary language. Add a file, another conversation, or a knowledge source only when it helps; Aerenium can also search the web and inspect the context in use.
Chat · Projects · Files · Web search · Context inspection · Temporary conversationsDeep Research can handle a longer investigation. Swarm can split a larger job among specialist workers while the main conversation stays focused on your goal.
Research runs · Source trails · Specialist model pools · Parallel workActivity shows the execution story in real time: plans, delegations, and tool calls. Open a delegated transcript, intervene when the direction is wrong, and keep approval gates on actions that deserve a human hand.
Live activity · Swarm transcripts · Intervention · Approval inbox · System logMove the result into Canvas as a document, code, diagram, or chart. Continue in Code Lab with workspace files and source control, or save the important parts as Highlights for the next conversation.
Canvas · Code editor · GitHub workflows · Highlights · Exportable conversationsInside Alpha 0.5
These are the real application surfaces, captured from one clean demonstration workspace. Open any image to inspect the full-resolution interface.
Memory you can inspect
The memory matrix separates how long information should last from when it should surface. Useful items can consolidate into longer-lived memory; repeatedly ignored items can be demoted instead of following you forever.
Open memory to search, edit, promote, or demote saved context. Use a temporary conversation when you want a clean slate, and mount knowledge sources only when they belong in the room.
Agency with boundaries
Browser access, desktop control, command policy, and delegated-agent budgets stay separate. Choose the boundary that fits the task, keep approval gates where you want them, and stop the work when you need to.
The in-app browser is sandboxed from your desktop. Its mouse, keyboard, downloads, access rules, and action history stay inside the browser workspace.
Set read and control permissions, scope allowed apps and domains, keep commands on Ask, and stop the agent when you need the keyboard back.
Bring your own model key, set ceilings for delegated agents, separate specialist model pools, and inspect usage and cost trends inside the workspace.
A complete workspace
Not a chatbot surrounded by novelty buttons. Each surface has a job, and each one can hand useful context or finished work to the next.
Work through an idea, search current sources, inspect what was read, and keep the conclusions attached to their evidence.
Carry decisions across sessions, search your own material, and choose the context that belongs in the room.
Turn chat output into durable documents, code, diagrams, and charts, then continue with files and source control.
Run bounded browser work, schedule saved tasks, and let alarms notify you while tasks hand a directive back to the assistant.
Split complex work across specialists, watch the execution story, open the transcripts, and intervene without losing the main thread.
Teach repeatable workflows with readable skill folders, connect external tool servers, and bring conversations in through Discord.
Start with the full workspace
Use Aerenium with your own model keys for $0. Paid plans add managed proxy credits and connected extras such as voice. Desktop automation and specialist agents are part of the app. Credit packs remain optional.
The short version
No. The Free tier runs the local workspace with your own model key, including computer actions and specialist agents. Paid plans add managed proxy credits and connected extras. You choose whether to use them.
No. The sandboxed browser and desktop Agent Mode are separate systems. Desktop read and control permissions are off until enabled, and app scope, domain scope, approval policy, and stop controls remain configurable.
Conversation history, memory, and workspace data are local by default. Content you deliberately send to a model or integration follows the provider you chose. Cloud backup is an optional, separate feature.
Yes. Activity exposes planning, delegation, and tool calls as they happen. Swarm keeps delegated transcripts available to inspect and steer, while the system log keeps a chronological operational record.
Download Aerenium free, bring your own model key, and try the loop, controls, and memory on work that matters to you. Add a paid plan only when you need the connected layer.
