SuperDev — Let your Agent troubleshoot like a developer.
Give your Agent one shared view of projects, services, logs, and deployments—local or remote. When logs run out, inspect the live process. Then verify the fix in the browser. Every real write is previewable, approvable, and auditable.
Local-first · MCP-ready · macOS + Windows desktop
Before and after SuperDev
Before: Copy logs by hand — Multiple tabs, lost lines, missing context
After: Read live and historical logs directly — One searchable stream with surrounding context
Before: Agent starts a shadow service — Different port, different config, different state
After: Share the same managed runtime state — Same process, same config, same evidence
Before: Humans verify every fix — Manual visits, screenshots, and back-and-forth
After: Verify in the browser and Network panel — Navigate, inspect, and return reproducible evidence
One workbench, from runtime to delivery
Troubleshooting is the signature workflow. The same runtime source of truth also manages every project, every log stream, and every delivery path.
Manage every project in one place
Projects and environments stay connected to the services that actually run across local and remote hosts.
- Local and remote hosts
- process / launchd / systemd / Docker
- Managed control and monitor-only
- One service and deployment model
Live logs, history, and search
Read the same stream in the desktop and the Agent, then move from one line to the exact surrounding context.
- Cross-service search and context lookup
- Rule filters and split panels
- Bookmarks and synchronized capture
- Duplicate folding and history
Breakpoints and browser verification
Use logs first. If evidence runs out, attach to the same managed process—then verify the result in the browser.
- Stack, scopes, and variables
- No shadow process or replacement PID
- Navigate, click, type, and capture
- Console and Network inspection
Pipelines, artifacts, and ingress
Operate the same delivery path your team trusts, from a validated DAG to a reversible release and durable ingress.
- Templates, variables, and artifacts
- Run history and log replay
- release/current rollback
- DNS, reverse proxy, and HTTPS
One real troubleshooting run
A single incident, followed from the user’s symptom to browser evidence. Choose any step to inspect the full context.
Start with the user’s real symptom
The user asks for an investigation. SuperDev does not invent or trigger an alert.
POST /api/payments returns 500 for a valid ¥698 payment. The account has ¥1,280. Please investigate and verify the fix.
Read the running service before touching code
payments-api is RUNNING; a real request still returns HTTP 500.
Use one breakpoint to inspect the missing values
The capture attaches to the same managed PID; it does not restart or replace the service.
The request amount is converted with the wrong multiplier. ¥698 becomes 698,000 cents instead of 69,800.
Fix the unit conversion in the repository
This is a normal Agent code edit—not a SuperDev-approved operation.
Preview the real write, then decide
restart_service is protected because it changes the managed runtime.
Return evidence from the user-facing path
The browser result and Network response agree: the payment now succeeds.
Real operations have hard boundaries
The safety gate is enforced by the local agent—not by a prompt asking the Agent to behave.
Every write moves through the same enforced path. Reads stay fast; real changes remain explicit and reviewable.
Preview
See the target, parameters, impact, and risk before anything changes.
User approval
Protected runtime writes wait for an explicit decision. No approval, no change.
One-time token
A short-lived token is bound to the approved operation and cannot be reused for another target.
Audit record
Preview, approval, execution, outcome, and context stay traceable after the operation.
All enforcement happens inside the local agent. No silent writes and no prompt-only safety boundary.
The questions a developer should ask
My Agent can already run shell and tail logs. Why do I need SuperDev?
Shell is useful, but it exposes commands—not one managed source of runtime truth. SuperDev connects projects, local and remote services, historical logs, pipelines, browser verification, approvals, and audit around the processes you already run. The Agent observes before acting, so it does not drift into a shadow service on another port or config.
Does my runtime data go to the cloud?
SuperDev is local-first. The desktop and MCP client talk to the local agent, and runtime data stays under your control. Remote hosts are connected only when you configure them; SuperDev does not require a hosted control plane for the core workflow.
Can an Agent operate production directly?
Reads can gather evidence. Protected writes such as restart or deploy move through preview, your approval, an operation-bound one-time token, and an audit record enforced by the local agent.
How do I install it, and which platforms are supported?
Desktop installers for macOS and Windows are available from GitHub Releases. You can also build from source. The project remains local-first and open source, with platform-specific capability notes documented publicly.
Local-first, transparent, and open source
Install the desktop app, connect an MCP-compatible Agent, and keep the runtime boundary on infrastructure you control.
Local-first
The desktop and MCP client share the local agent. Your core runtime workflow does not depend on a hosted control plane.
Debug language support
Default: Go, Python, Rust, C/C++. Experimental: Node (bundled adapter) and Java/Kotlin (self-provided adapter).
Desktop installers
Download native desktop installers for macOS and Windows from GitHub Releases.
Open source & source build
Apache-2.0 licensed. Inspect the code, follow the build docs, and run the entire stack from source.
Documentation · 中文文档 · GitHub · Downloads