Reforception puts the work above the model. Plan, chat, notes, agents, repositories, tools, and connected machines stay in one operating environment while you choose the intelligence and the authority for each job.
You choose the intelligence. Reforception coordinates the work. You stay in control.
Reforming Perception
Reforception workspace
One request · coordinated execution
Request
“Review the auth change, repair anything serious, run the checks, and show me what changed.”
Plan
Shape the work
Break the outcome into review, repair, checks, and approval.
Intelligence
Selected model
Provider and model stay explicit.
Tools
GitHub + workspace
Only connected resources are available.
Agent execution
Bounded work in progress
Result
Evidence before approval
Changes, checks, failures, and remaining decisions return to the same workspace.
From request to result
The point is not to collect AI models in one screen. The point is to move real work through the right intelligence, tools, permissions, and execution environment without losing the thread.
01
Start with the result you want, not a tool-by-tool workflow you have to wire together yourself.
02
Turn the request into a plan you can read, adjust, and approve before execution begins.
03
Use the connected provider, model, or agent that fits the job. Reforception does not silently swap your choice.
04
Let bounded agents work with the repositories, services, and machines you explicitly connected.
05
See what happened, what passed, what failed, and what still needs your decision before work moves forward.
Why Reforception
Models will change. Providers will change. The workspace around your goals, decisions, permissions, and results should not have to.
01
Plans, notes, and the shape of the work stay put when you change models. The model is the intelligence for a task, not the place the work lives.
02
Choose the provider and the model for the job. If that choice cannot run, the work stops and says so. It does not quietly continue on a different provider.
03
Local work and remote work are different places. Reforception is built so you can tell where a task is running, and what it is allowed to touch.
One product
Chat, Plan, Build, Notes, Connections, and Host Agent are not separate products. They are different ways into the same workspace, with the same work and boundaries behind them.
Ask, compare, and revise in a thread that stays attached to the project instead of disappearing into a disposable chat tab.
Turn a loose request into a sequence you can check. Planning stays inside the workspace. It is not a separate product.
When a task is clear enough to run, pass it to an agent and watch the progress. You see status and a result, not a silent black box.
Keep drafts, decisions, and reference material in the same workspace as the conversation, instead of in a second app you forget to update.
Bring in outside services on purpose. A connection is something you add. It is not assumed, and it is not hidden.
Some jobs belong on a machine you control. A Host Agent is the boundary for that local or remote execution.
The product in action
These are real Reforception product surfaces rendered from the application components with safe fixture data. The point is not six separate tools. It is one system that keeps the plan, execution, evidence, and people in the same place.
Plan
Reforception Plan turns a project into visible work: what is active, what is blocked, what is ready, and what still needs a decision. The AI planning assistant can propose changes without silently rewriting the plan behind you.
Build
Build is conversation-first. Connect a repository, describe the outcome in ordinary language, choose an eligible worker, then follow the agent activity, changes, and checks without turning the product into an IDE you have to learn first.
Agents
The Agents surface makes workers visible as part of the system. You can see who is working, what is assigned, what has run, and whether a worker is allowed, requires approval, or is not permitted for a project.
Intelligence
Hosted models and models running closer to your own machines can share one working environment. Select the provider and model for the task in front of you instead of rebuilding the job inside a different app.
If the selected model cannot run, Reforception is designed to stop and tell you rather than quietly continue through a different provider or billing path.
Connected compute
A desktop, home server, laptop, or remote host can become a bounded execution node through the Host Agent. Reforception keeps the machine, project, and authority visible so using your own compute does not mean granting an agent the rest of the box.
A Host Agent can run bounded work near the files and resources a job needs, on hardware you already control.
Remote work can run in a separate connected environment with its own permissions, identity, and limits.
Control
Reforception is built around visible choices: what is connected, which model is selected, where work runs, and what an agent is actually allowed to do.
Services show up because you connected them. You can see the list.
The provider and model for a task are a choice, not a background swap.
An agent does not receive a blank check to a machine. Access is scoped to what the job is allowed to do.
Local execution and remote execution are labeled. Credentials for the application stay server-side, not in this public website.
Example authority
A worker can be able to read a repository and work inside an isolated workspace without automatically gaining permission to publish, merge, deploy, or reach the rest of a machine.
Read repository
Inspect the bound project context.
Work in isolated workspace
Make bounded changes away from the published source.
Run approved checks
Use only the command kinds the workflow grants.
Publish or deploy
Not implied by repair or Build authority.
Triage
Reforception Triage is being built as a review-and-repair pipeline that separates the first scan, independent confirmation, repair authority, and verification. It is not presented here as a finished private-beta feature yet.
01
ReviewFast, bounded review for suspicious changes and risk hotspots.
02
ConfirmA stronger reviewer independently confirms or rejects important findings.
03
WorkAn eligible Build worker receives only approved findings and bounded context.
04
ProveChecks and an independent review decide whether the repair is actually clean.
Private beta
Reforception is actively being built. If you already have access, open the application and continue your workspace there.
