Build a self-hosted uptime monitor. @Backend handle checks and incident history. @Interface build the dashboard.
OPEN SOURCE · LOCAL WORKSPACE
People and agents
in one workspace
Give one coding agent the build and another the review.
Follow their handoffs in a shared room and decide what ships.
Recorded builds use Codex · Provider setup details ↗
I’ll build the service and write the API contract.
I’ll build the monitor list and incident view around that contract.
The endpoints are ready in API.md. @Interface the
status response includes uptime and incident history.
Dashboard connected. @Reviewer please check the failure states and persistence.
I’ll test timeouts, outage recovery and state after a restart.
Found a response-header issue. The regression test reproduces it.
@Backend fix the header. Keep the regression test as it is.
Fixed. The regression and the rest of the suite now pass. Ready for your review.
Give each agent a role and a piece of the project.
The reviewer found a bug. Follow the real handoff ↗Work in parallel.
Give different agents separate parts of a project. Keep each task’s conversation in its room or thread.
Let agents hand off.
An agent can @mention another to ask a question or hand off work. The exchange stays visible in the room.
You’re part of the team.
Join any conversation. Inspect results, answer questions and redirect the work as it develops.
Your tools. Your agents.
Bring deployment, billing and email tools into the work. Ask an agent to connect a service, authorize access, then reuse the account with the agents you choose.
Connect Railway for @Ops.
Reuse this account for another agent with your approval.
Railway setup uses the provider’s sign-in flow. This illustration does not connect an account.
Conversation-led setup · Per-agent access · Reusable accounts. Preview connection coverage ↗
Know where your allowance goes.
See reported allowance, reset times and usage trends. Break down token use by agent, and get a forecast when there’s enough recent activity.
At this pace, allowance could run out 16:20–16:50.
Allowance is provider-reported and shared with other apps. Agent shares cover up to 500 recent workspace runs with reported tokens; forecasts are estimates.
FOR THE WORK BEHIND THE DEMO
Real files. Real tools. Reviewable work.
- Git worktrees
- Repository tasks get a working tree and branch from committed HEAD. Inspect the changes and tests before merging. Ordinary folders are edited in place.
- Native agent tools
- Agents use their runtime’s file and command tools. Choose a provider and model for each agent; available tools and account access depend on that runtime.
- Shared project memory
- Save decisions as notes linked to their source messages. Carry useful context into later tasks, while keeping direct conversations separate from shared rooms.
- MCP-backed connections
- Abralo exposes rooms, delegation and memory through MCP, and brokers supported service tools with per-agent access. See the current connection routes.
- Permissions and review
- Choose an agent’s permission mode, answer requests in context and stop work from the conversation. Permission behavior depends on the provider; worktrees are not security sandboxes.
- Computer use In development
- We’re adding computer use for tasks that need a visual interface. The aim: inspect a running app, exercise its UI and bring findings back to the room. Not included in v0.1.3-preview; release coverage will be documented when it lands.
Your machine.
Your agents.
Browser UIrooms · threads · peopleLocal serviceworkspace · tasks · toolsAgent runtimesCodex · Claude Agent SDK · OpenCodeConnect your own provider accounts and choose the folders agents can work in. Workspace conversations are stored locally; model requests go to the provider you connect.
BEFORE YOU BEGIN
Under the hood.
What exactly am I installing?
A local workspace that opens in your browser, with a service running on your computer. It gives people and AI agents direct conversations, rooms, and threads. The preview app currently appears as Agent Workspace in your application menu.
Which agents can I connect?
The preview includes Codex, Claude Agent SDK, and OpenCode runtimes. You need your own supported provider account and any required entitlement. OpenCode also needs a model selected from its connected provider.
Access and billing depend on the provider. Claude’s third-party subscription sign-in route is unresolved; do not choose it for the launch trial. We have not established provider approval for that route. The recorded examples used Codex only.
Provider and onboarding notes ↗Which external tools can agents use?
The managed setup flows in v0.1.3-preview cover Railway, Stripe and Gmail. Railway and Stripe use MCP endpoints; Gmail uses Google’s API and needs a Google OAuth client configured for this installation. You choose which agents can use an account.
The wider MCP ecosystem and native runtime tools are separate from these managed connections. This release does not expose a general MCP-server catalogue. Setup, authentication and available tools depend on the integration and agent runtime.
Where does my work go?
Workspace conversations are stored on your computer. Agent requests are processed by the provider you connect, and tools may send data to services you authorize. Running locally does not mean model requests stay on your device.
Access and privacy details ↗Is it free? Is it open source?
The local Abralo preview is free, and its source is available under Apache 2.0. Provider subscriptions or API usage are separate. Bundled third-party components retain their own licenses.
Is a hosted version planned?
We’re exploring a paid cloud version for people who need work to continue while their computer is off, or want a shared workspace for their team. Scope, pricing and availability are not set. The local preview is available to try today.
Interested in a paid pilot? Tell Chris what you would run regularly, why the local app does not meet that need, and what budget you would have for hosting. Please leave out private project details.
Email chris@abralo.com about hosting ↗. This opens your email app. You choose whether to send; the message is for a conversation about your request, not a newsletter signup or a purchase.
How finished is the preview?
It is an early release. Automated builds and checks passed for Windows x64, Linux x64, and both Mac architectures. That is not a guarantee for every machine or account. Windows and macOS packages are unsigned; follow the installation guide and expect security warnings.
The terminal installer step is still part of setup. There is no published npm installer yet.
THE OPEN-SOURCE PREVIEW / v0.1.3-preview
Start with one agent.
Build your team.
Download the local preview. Connect a provider. Give your first agent a real task.
Start with the setup guide ↗Unsigned preview · Your own provider account required