installer
Deep integration
Guided setup. Hooks, region and rule base wired for you on the first run.
VibeDefendLive · every plan · EU + US
AppSec lives inside your agent. Every prompt.
Not a scanner. A security guard that learns your business rules and watches every conversation. Every prompt reviewed in real time. Memory that does not reset. Autopilot by default.
Native on
One install. Every coding agent.
Coverage
Two ways to come on board. Five deep, every other through MCP.
Claude Code, Cursor, Windsurf, GitHub Copilot and OpenAI Codex get a guided setup that takes care of everything. Every other MCP-compatible agent plugs in over the open protocol.
bring-your-own config
MCP support
Point your agent at VibeDefend's MCP endpoint and the agent talks to it over the open protocol.
How it works
Five steps, from first scan to every prompt.
VibeDefend reads your repo once, plugs into your agent, and stays in the loop. No babysitting, no rulebooks to write by hand.
01 · FIRST SCAN We read your repo and extract your rules
Auth flows, multi-tenant scopes, money paths, idempotency, audit trails, role checks. VibeDefend learns what your codebase already enforces and keeps it in your private base. Region-pinned, tenant-isolated, no training.
first scanscanningrule families extracted
from repository00rule families · per repo02 · CONNECT Pick your region, pick your agent
Pick your region (EU or US) and your agent. Claude Code, Cursor, Windsurf, GitHub Copilot and OpenAI Codex get the deep integration that wires everything for you. Every other MCP-compatible client connects in a couple of clicks.
connect your agent10 agents · two regionsregionEUUSnative installerclaudecursorwindsurfcopilotcodexmcp compatiblegeminiclinecontinuezedantigravity10agents supported today03 · REVIEW Every prompt, reviewed in real time
When you prompt the agent, VibeDefend steps in. It surfaces the rules that matter for the task. The agent decides if your draft respects them and flags violations before writing any code.
every prompt, reviewedliveuser promptreviewrules base✓rules metverdictevery prompt sent to the agent
is checked against repo rules∞prompts protected · session04 · MINE Every few prompts, we re-mine
Every few prompts (three by default, you choose), VibeDefend captures the new rules that emerged from the conversation. Autopilot adds them to your base. Manual mode queues them for your review. Your rules grow with your codebase.
continuous rule learning3 / 3 promptsprompt stream›add a refund endpoint›handle partial refunds›log to audit tablerule base043default prompt cadence05 · PERSIST Memory that does not reset
Long sessions, conversations that restart, agents that swap mid-flow, your security context holds. The moment the next prompt comes in, the right rules are already there. Nothing gets lost.
memory that holdssession activeconversation contexthistorysession startnowsession.window100%vibedefend rulespersisted · 03auth.requireOwneridempotency.keyaudit.trail.complete0rules lost · ever
Autopilot vs Manual
Default is autopilot. Manual is one flag away.
VibeDefend runs hands-off out of the box. Need a check on every new rule? Flip to manual and the queue lands in your inbox.
default
Autopilot
compliance-grade
Manual
Who proposes new rules
VibeDefend learns them
You add them
Cadence
Every few prompts
On demand
Validation
Added straight to your base
Queued for review
Visibility
Live diff in the CLI
Dashboard inbox
Best for
Solo and small teams
Regulated, audit-heavy
Privacy by design
Your code stays yours. EU or US, never both.
VibeDefend reads in-region, retains nothing at rest, and never trains a model on your repo. Rule extraction is per-tenant and ephemeral.
Read in-region
Repo content stays inside the region you pick. EU endpoints serve EU tenants only, US endpoints US only. No cross-region traffic without an explicit opt-in.
Zero retention, zero training
VibeDefend does not store your repo at rest. It does not train any model on your code. Extracted rules live in your private base, tenant-isolated, ephemeral on demand.
Audit trail by default
Every rule mining pass, every hook decision, every auto-add lands in your audit log signed by the agent identity. Provable retroactively.
FAQ
Quick questions before you flip it on.
How do I install VibeDefend?
Pick your region (EU or US) and the agent you use most. For Claude Code, Cursor, Windsurf, GitHub Copilot and OpenAI Codex, the guided setup handles everything for you. For any other MCP-compatible client (Gemini, Cline, Continue, Zed, Antigravity and more), VibeDefend ships as a Model Context Protocol server so any compatible client picks it up automatically.
What gets learned on the first scan?
The business-logic rules your codebase already enforces. Auth flows, tenant scopes, money paths, idempotency, audit trails, role and permission checks, refund caps. A production repo typically yields 20 to 40 distinct rule families on the first pass. No setup, no prompt engineering.
What happens on every prompt?
VibeDefend steps in between you and the agent. The rules that matter for the task surface in the agent's working context. The agent decides if the draft respects them and flags violations before writing any code. Verdicts surface inline.
What about long sessions and resets?
Your security context lives outside the conversation. Long sessions, restarted chats, swapped agents, restarted IDEs, the rules stay live. The moment the next prompt comes in, VibeDefend has the right rules ready.
Does it work without CybeDefend SAST or SCA?
Yes. VibeDefend runs standalone with just an agent and a repo. The CybeDefend scanners (SAST, SCA, IaC, Container, Secrets, CI/CD) pair well with VibeDefend but are not required.
Get started
Install in your AI agent. First scan in 5 minutes.
No credit card. No setup call. Pick your agent, paste the command, and Cybe enforces your rules from the very next prompt.
Region
claude mcp add cybedefend --transport http https://mcp-eu.cybedefend.com/mcpHosted by us, no install. Just point your agent at the VibeDefend endpoint.