01
CONTROLLED TOOLS
PMCAgent exposes narrowly scoped operations instead of unrestricted shell access. Each capability is designed around an approved action, workspace or service.
← Back to PMC AI ProductsEngineering context.
CONTROLLED
CUSTOM AI DEVELOPMENT AGENT
Engineering context.
Controlled execution.
PMCAgent is PMC AI's customized development agent for complex, long-running software projects. It combines persistent project memory, controlled VPS tools, diagnostics, deployment workflows and verification into one engineering operating layer.
PMCAgent Control Plane
Purpose-built for real projects
64
controlled capabilities
/opt
approved project root
LAST-GOOD
runtime recovery
MEMORY
persistent checkpoints
The design principle is simple: give the agent enough capability to complete serious engineering work, but expose that capability through narrowly defined, verifiable operations instead of unrestricted server access.
System architecture
Designed as an engineering control layer.
PMCAgent is not a single automation script. It is a structured layer between an AI engineering assistant and the project infrastructure, with explicit controls for state, execution, validation and recovery.
02
PROJECT MEMORY
Persistent project context preserves verified architecture, workflow rules, checkpoints and operational practices so long-running development can continue from known state.
03
FAIL-SAFE RUNTIME
The agent runtime is protected by an active-runtime and last-good model, allowing the launcher to recover from an invalid candidate instead of losing the management interface.
04
VERIFICATION FIRST
Builds, type checks, service status, browser QA, logs and screenshots are used as evidence before a change is considered complete.
Engineering lifecycle
From request to verified outcome.
The workflow is intentionally systematic so a complex project can be changed safely without skipping investigation, protection or real-world verification.
01
UNDERSTAND
Load project memory and verify the current architecture, previous decisions and the exact workspace involved.
02
INVESTIGATE
Inspect relevant source, logs, services and runtime state before changing production behavior.
03
PROTECT
Create a targeted backup or recovery point for the components that will actually be modified.
04
CHANGE
Apply the smallest controlled modification needed through an allowlisted tool or project operation.
05
VALIDATE
Run syntax, type, build, configuration or workflow validation appropriate to the changed component.
06
DEPLOY
Restart or reload only the required application or service after validation has succeeded.
07
VERIFY
Confirm the real outcome using status checks, logs, browser QA and visual evidence where applicable.
08
CONTINUE
Record important verified state so the next engineering session starts from the latest confirmed checkpoint.
Controlled capability
Broad enough for development. Narrow enough for control.
01
Workspace Engineering
Read, search, create, update, move and validate project files inside approved development workspaces under /opt while keeping operations auditable and bounded.
02
Build & Deployment
Run approved TypeScript, npm, Prisma and deployment workflows, create backups, validate changes and restart only the services required by the release.
03
Service Operations
Inspect and manage allowlisted PMC services, PM2 applications, Nginx, systemd, scheduled jobs, network state and storage without exposing unrestricted VPS control.
04
Application Diagnostics
Analyze logs, process state, database health, permissions, disk usage and production incidents using read-only diagnostics before deciding whether a code or infrastructure change is necessary.
05
Browser & Visual QA
Inspect live PMC pages, run controlled browser interactions and capture visual proof so UI changes can be checked against the real rendered application rather than source code alone.
06
Specialized Automation
Extend the agent with purpose-built controls for systems such as n8n workflows, YouTube Studio cover operations, mail infrastructure and DNS diagnostics.
07
Recovery & Rollback
Use protected backups, last-good runtime behavior and narrow rollback paths to reduce the risk of losing access during changes to critical development infrastructure.
08
Persistent Engineering Context
Store verified project decisions and checkpoints so recurring development sessions do not repeatedly rediscover architecture or accidentally reintroduce obsolete approaches.
09
Evidence-based Completion
A task is not treated as complete merely because code was written. PMCAgent supports validation through compilation, service health, logs, browser inspection and screenshot evidence.
Safety & resilience
Built to reduce operational mistakes.
PMCAgent's value is not only what it can do, but how those actions are constrained, validated and recoverable when working on a live engineering environment.
✓
Approved workspaces and services instead of unrestricted server-wide mutation
✓
Read-only diagnostics separated from write operations
✓
Secret-aware tools that avoid returning credentials or private keys
✓
Validation before restart or deployment for critical components
✓
Targeted backups and explicit rollback paths
✓
Last-good runtime protection for the MCP management layer
✓
Visual and operational verification before claiming success
✓
Project memory to prevent repeated or obsolete fixes
Where it fits
For projects where continuity matters.
PMCAgent is designed for engineering work that spans many components, repeated sessions and operational checkpoints—not just one-off code generation.
Large application development
Coordinate changes across frontend, APIs, databases, workflow automation and infrastructure while preserving project context over many sessions.
Production incident analysis
Investigate failures with logs, health checks and runtime diagnostics before applying a narrow corrective change.
VPS operations
Manage approved services, Nginx, deployment state, mail infrastructure and system diagnostics through controlled interfaces.
Workflow automation projects
Inspect, back up, validate and update complex n8n workflows without relying on ad-hoc manual edits.
Release verification
Combine build checks, service status, browser testing and screenshots to verify that a requested change is actually live.
Long-running engineering programs
Carry architecture, checkpoints and operational rules forward so the AI assistant behaves like a continuous project engineering partner rather than a stateless chat.
CUSTOM PMCAgent
Build an AI development agent around your real engineering environment.
The PMCAgent model can be adapted to the workspaces, services, deployment controls, diagnostics and verification requirements of a complex software project.