Catch the mistakes AI tends to make
AI can generate working-looking code that leaks secrets, skips auth, builds unsafe queries, misses rate limits, or executes untrusted model output. VibeLint checks for those patterns early.
AI agent security platform
Secure every action your AI agents take.
VibeLint checks code writes, tool calls, and workflow actions before they become risk. Block unsafe generated code, gate sensitive actions, and keep a searchable audit trail across IDEs, workflow builders, agent frameworks, and custom runtimes.
AI agent securityCode scanningAgent permissionsAction logsApproval workflowsMCP & HTTP API
01 / What is VibeLint
A control layer for what AI writes, runs, and touches.
VibeLint helps developers and teams secure AI-assisted development by checking generated code, agent tool calls, and workflow actions before they turn into real risk.
In plain English
VibeLint is a security layer for AI coding agents. It scans AI-generated code, detects risky patterns, and helps developers prevent insecure code and unsafe agent actions from reaching real projects.
One-line positioning
VibeLint helps you secure what AI writes, runs, and touches.
What it gives you
- Catch risky AI-generated code before it lands in your project.
- Decide when agent actions should run, stop, or wait for approval.
- See what agents did through searchable action logs.
- Use it with IDE agents, MCP-compatible tools, workflow builders, frameworks, and custom agents.
Code scanning flow:
AI Agent -> Code Change -> VibeLint Scan -> Risk Detection -> Allow / Warn / Block / LogAgent action flow:
AI Agent -> Tool Call or Action -> VibeLint Permission Check -> Allow / Block / Require Approval -> Log Evidence02 / Why use it
VibeLint is useful when AI is no longer just suggesting code.
Once agents can write files, call tools, create records, send messages, or change production-adjacent systems, you need a security layer that checks actions before they happen and leaves evidence afterward.
Control agents before they touch sensitive systems
When agents call tools, send messages, update records, or change files, VibeLint can return allow, block, or require approval decisions.
Keep evidence you can review later
Action logs record the agent, tool, action, status, risk level, approval state, and redacted context so you are not guessing what happened after the fact.
Know whether upgrading is worth it
Free is enough to try local code checks. Pro is useful when you rely on AI coding daily and want the full detector suite, logs, permissions, approvals, and dashboard history.
How it compares to tools you may already use
Traditional SAST tools
Traditional scanners usually review repositories, pull requests, or dependencies after code exists. VibeLint focuses on the moment AI tries to create, modify, or act on code.
Dependency scanners
Dependency tools focus on packages. VibeLint includes dependency risk as one part of a broader AI agent security loop.
Manual review
Manual code review is still important. VibeLint catches common AI-generated risks earlier so reviewers spend less time on preventable mistakes.
AI coding assistants
Coding assistants generate code. VibeLint checks the generated code and agent actions for risk before they land.
03 / Who it is for
Built for builders letting AI touch real code and tools.
VibeLint is useful whenever an AI system can produce code, modify files, call APIs, use tools, or run a workflow that affects a project or customer.
Solo developers
Developers using AI coding assistants who want a fast security check before generated code enters their project.
Startups
Small teams shipping quickly with AI tools that need visibility into secrets, auth mistakes, injection risk, and risky agent activity.
Teams using AI agents
Teams allowing agents to edit files, call tools, touch APIs, update data, or run workflow actions in real projects.
Security-conscious builders
Builders who want lightweight AI security without rolling out a heavy enterprise security stack before the product is ready.
Vibe coding safely without ignoring security.
Preventing secret leaks from generated code.
Reviewing AI-generated authentication and database code.
Controlling agents before they send messages, change data, call APIs, or touch sensitive files.
04 / How it works
One security loop around agent work.
The core idea is simple: check the risky thing before it becomes a real change, then record enough evidence to understand what happened later.
Agent proposes work
A coding agent, workflow agent, or custom runtime prepares to write code, call a tool, or take an action.
VibeLint checks risk
VibeLint scans generated code, checks permission policy, scores action risk, and looks for detector matches.
Decision is returned
The action can be allowed, warned, blocked, logged, or gated behind approval depending on the plan and policy.
Evidence is recorded
Scan metadata, issue summaries, permission decisions, action status, risk level, and redacted context become reviewable history.
05 / Setup
Set up code scanning via MCP, or connect agents via HTTP API.
VibeLint supports two integration paths: MCP for IDE code scanning and HTTP API for agent permissions, logs, and approvals from any runtime.
Path A: IDE code scanning (MCP)
- 1Open the dashboard, choose your editor and operating system, then copy the generated setup command.
- 2Run the command and approve the named device in your signed-in browser session.
- 3Browser approved — waiting for this device to finish setup
- 4Wait for “Installation verified,” then fully restart the editor and open a new agent task.
- 5Wait for Scanner ready from the real IDE MCP session. Only that session completes setup.
Path B: Agent API (HTTP)
- 1Create a VibeLint account and open the dashboard.
- 2Go to Agent Identities and create a new agent.
- 3Copy the agent key shown after creation.
- 4Use the agent key to call the VibeLint HTTP API from your agent runtime.
- 5Call POST /api/agent-permissions/check before sensitive actions.
- 6Call POST /api/agent-actions/log after each action to record evidence.
Dashboard command
Browser approval
Verified installation
Restart editor / open a new agent task
Scanner readyWhere the exact commands live
The dashboard generates one short-lived setup command for your account, plan, editor, and operating system. Code Security setup uses browser authorization. The command installs a private side-by-side runtime, safely changes only the VibeLint MCP entry, preserves other servers, and runs a local scanner probe.
IDE & MCP agents
Codex
Guided setup preserves unrelated MCP servers and updates only mcp_servers.vibelint.
Cursor
Guided setup preserves unrelated MCP servers and updates only mcpServers.vibelint.
Claude Code
Guided setup preserves unrelated MCP servers and updates only mcpServers.vibelint.
Antigravity
Guided setup preserves unrelated MCP servers and updates only mcpServers.vibelint.
Devin Desktop (formerly Windsurf)
Guided setup preserves unrelated MCP servers and updates only mcpServers.vibelint.
Setup recovery
staged_not_authorizedThe candidate runtime is staged but browser authorization is unfinished.
Return to the browser approval page, then rerun the same setup command while it remains valid.
authorization_exchange_unavailableThe browser approved this device, but credential exchange is temporarily unavailable.
Leave setup open; it will retry safely while the device code remains valid.
credential_revokedThis device authorization was revoked.
Generate a fresh dashboard setup command and approve this device again.
runtime_variant_mismatchThe staged runtime variant does not match the required account tier.
Generate a fresh setup command so the correct signed runtime variant can be installed.
path_runtime_mismatchA terminal launcher resolves to a runtime other than the active managed runtime.
Run guided repair so VibeLint can replace only a proven owned launcher with the stable dispatcher.
protected_legacy_installA proven older VibeLint launcher is in a protected system location.
Keep local protection active and follow the single elevated cleanup action shown by setup.
local_ready_cloud_sync_pendingLocal protection is ready while dashboard acknowledgement is pending.
Keep the editor open and retry acknowledgement; local protection remains active.
editor_restart_requiredInstallation is verified, but the editor has not started a new MCP session.
Fully restart the selected editor and open a new agent task.
deep_engine_warmingThe local gate is ready while deep analysis is warming.
Continue using local protection and retry deep status after initialization completes.
gate_detector_failedThe effective local gate detector or worker failed its health check.
Run doctor repair once; if the failure remains, reinstall the signed runtime and report the correlation code.
Workflow builders & agent frameworks
n8n
Use HTTP nodes to call VibeLint permission checks and log agent actions.
Zapier Agents
Call VibeLint API from Zapier actions for permission and logging.
Make
Use HTTP modules to connect AI agent workflows to VibeLint.
Dify
Add VibeLint API calls as tool nodes in Dify workflows.
Flowise
Connect Flowise agents to VibeLint via HTTP tool nodes.
LangChain / LangGraph
Add VibeLint permission checks as tool calls in LangGraph agents.
CrewAI
Wrap VibeLint API calls as CrewAI tools for agent security.
OpenAI Agents SDK
Register VibeLint as a function tool in the OpenAI Agents SDK.
Custom agents
Any runtime that can call HTTP can use VibeLint for permissions, logs, and approvals.
06 / Core features
The product has three pillars: code security, logs, and permissions.
Together these cover the most important risk surfaces in AI-assisted development: what the agent writes, what it tries to do, and what evidence remains.
Available now
Real-time AI code scanning
VibeLint scans AI-generated or AI-edited code for risky patterns before the code becomes part of a project.
- Designed for pre-write checks in IDE and MCP workflows.
- Works with manual scans such as vibelint scan . when a deeper pass is needed.
- Returns issue type, severity, line context, description, and fix guidance.
Available now
Security rule detection
The detector library covers common mistakes introduced by AI coding tools, including secrets, injection, auth, config, AI-specific risk, and dependencies.
- Free includes the limited detector set.
- Pro and Max include the full detector suite.
- Detector results can be reviewed by category and severity.
Available now
MCP integration
VibeLint can run as an MCP server so compatible coding agents can call it during code generation and review.
- MCP lets an AI agent call tools outside the chat window.
- VibeLint exposes a scan and security-check loop through that workflow.
- The dashboard generates one guided setup command for your editor and OS.
Available now
Agent identities and API keys
Teams can create named agent identities and scoped agent keys for runtimes that send action logs or permission checks.
- Each agent can represent an IDE assistant, internal bot, workflow, or custom runtime.
- Keys are shown once and can be revoked.
- Agent limits depend on the selected plan.
Available now
Agent action logs
Action logs show what AI agents attempted or executed, including tool, action, status, risk level, timing, rollback reference, and redacted metadata.
- Useful for auditing agent behavior across workflows.
- Supports statuses such as attempted, executed, failed, blocked, gated, and rolled_back.
- Logs can be filtered by agent, tool, status, risk, session, and date.
Available now
Permission manager
Permission policies define what agents may do, what should be blocked, and what requires human approval first.
- Rules can match tools, actions, environments, agents, and structured conditions.
- Decisions include allow, block, and require_approval.
- Approval queues are available on paid plans that include them.
Available now
Dashboard visibility
The dashboard helps review scans, projects, detectors, agents, action logs, policy decisions, approvals, plan limits, and setup state.
- Scan history is grouped by project.
- Detector breakdowns help prioritize fixes.
- Plan limits are applied automatically by account plan.
Available now
Privacy-first scan model
Code scanning runs locally by default. Dashboard sync focuses on metadata and summaries rather than raw source files.
- Scanner findings can sync as security metadata instead of full source files.
- Agent payload logging masks common sensitive key names.
- Teams decide what agent context they send to the action log API.
07 / Security rules
Detection coverage is organized by detector tier.
Free includes the starter detector set. Pro and Max include the full current detector suite for deeper coverage across AI-specific risks, dependencies, auth, rate limits, and more.
Hard-Coded Secrets
API keys, tokens, passwords, private keys, and credentials embedded directly in source code.
Why it matters
Secrets in code can leak through commits, logs, screenshots, bundles, or copied snippets.
VibeLint behavior
Warns or blocks depending on plan, policy, and workflow configuration.
Safer alternative
Move secrets to environment variables or a secrets manager, then reference them through safe configuration.
Insecure Authentication
Weak auth checks, unsafe password handling, unverified sessions, or patterns that make authentication easy to bypass.
Why it matters
AI-generated auth code can look complete while missing the actual security checks that protect user data.
VibeLint behavior
Flags risky auth patterns and provides fix hints for safer validation.
Safer alternative
Use verified sessions, strong password hashing, explicit permission checks, and trusted auth helpers.
Injection Risk
Parser-backed flows from untrusted input into SQL/NoSQL, shells, HTML, files, outbound requests, templates, XML parsers, deserializers, and dynamic object writes.
Why it matters
Injection bugs appear when external values gain control over code, queries, destinations, paths, or executable rendering contexts.
VibeLint behavior
Blocks proven unsanitized flows and warns when cross-file source or sanitizer resolution is incomplete.
Safer alternative
Use parameterized and typed APIs, context-specific sanitizers, strict allowlists, canonical containment, and hardened parsers.
Framework Configuration Analysis V2
Parser-backed debug and error exposure, proxy trust, CSRF, template, management endpoint, static-content, security middleware, request-limit, and timeout policy issues.
Why it matters
Development and framework-default configuration can silently expose source, diagnostics, trusted headers, or sensitive runtime controls in production.
VibeLint behavior
Blocks explicit high-confidence unsafe production controls and warns on development-only settings or project-level missing hardening.
Safer alternative
Separate deployment profiles, minimize public diagnostics, validate proxy boundaries, and use framework or proxy hardening controls.
CORS Policy Analysis V2
Credentialed wildcard or reflected origins, weak validators, null origins, broad preflight rules, middleware ordering, and infrastructure overrides.
Why it matters
Unsafe CORS trust can expose authenticated browser responses to an attacker-controlled origin.
VibeLint behavior
Blocks proven credentialed arbitrary-origin policies while public non-credentialed wildcards remain review warnings.
Safer alternative
Use exact HTTPS allowlists, strict parsed-origin checks, and Vary: Origin for dynamic responses.
Static Analysis Findings
Curated offline Semgrep rules for cryptography, TLS, randomness, filesystem safety, sensitive logging, ReDoS, memory safety, containers, infrastructure, and CI supply-chain configuration.
Why it matters
Semantic static patterns catch dangerous APIs and insecure configurations that specialized flow detectors do not own.
VibeLint behavior
Blocks reviewed high-confidence findings, warns on audit rules, and fails closed if the local engine cannot complete.
Safer alternative
Use the rule-specific fix guidance, correct the code, and resubmit the complete file for scanning.
Prompt Injection Trust Analysis V2
Untrusted user, RAG, web, file, memory, tool, MCP, and multimodal content crossing into privileged instructions or capable-agent context.
Why it matters
Prompt injection becomes dangerous when untrusted content can influence sensitive context, memory, tools, or privileged actions.
VibeLint behavior
Keeps ordinary user-role chat clean, warns on unresolved indirect flows, and blocks proven privileged trust-boundary violations.
Safer alternative
Separate instructions from data, restrict tools, authorize actions outside the model, and require approval for high-risk operations.
LLM Output Execution V2
Model-generated code, commands, dynamic modules, written scripts, client-tool calls, and MCP dispatch reaching local execution or privileged actions.
Why it matters
Generated output and tool calls are untrusted decisions; schema-valid arguments can still trigger destructive or unauthorized operations.
VibeLint behavior
Traces provider-proven model output across statements and helpers, blocks host execution or unrestricted privileged dispatch, and warns when validation, approval, authorization, or sandbox controls cannot be proven.
Safer alternative
Use fixed dispatch maps, strict schemas, application authorization or exact-action approval, and proven isolated runtimes for generated code.
Vulnerable Dependencies
Known vulnerable packages or dependency patterns found in project manifests.
Why it matters
AI agents often install packages quickly without checking the security history of what they add.
VibeLint behavior
Reports known dependency risk when available to the scanner.
Safer alternative
Upgrade, pin safe versions, replace abandoned packages, and review new dependencies before merge.
Rate Limiting & Abuse Controls V2
Missing, disabled, bypassed, unattached, misordered, local-only, or fail-open request limits; costly operations; GraphQL fan-out; and WebSocket message abuse.
Why it matters
Authentication, paid-provider, notification, payment, export, and realtime endpoints can become brute-force or resource-exhaustion paths.
VibeLint behavior
Warns when single-file context is incomplete and blocks project-proven high-risk endpoints without effective frequency, quota, cost, or message controls.
Safer alternative
Use trusted composite partitions, shared counters, burst plus sustained limits, and charge quota before expensive work.
Missing Authentication
Endpoints or routes that appear to expose sensitive data or actions without authentication.
Why it matters
AI can generate complete-looking APIs while forgetting the guard that proves who is calling.
VibeLint behavior
Flags routes where auth appears absent or too weak.
Safer alternative
Require authenticated users, service tokens, roles, or explicit permission checks before sensitive work.
08 / Policies, actions, and audit trails
Policy decisions make agent behavior reviewable.
VibeLint can treat agent work as structured decisions: the agent, tool, action, risk, policy result, approval state, and redacted context are separated so teams can understand what happened.
Policy examples
- Block all hardcoded secrets.
- Warn on missing rate limiting.
- Require approval before modifying auth-sensitive files.
- Require approval before database schema changes.
- Block destructive shell commands.
- Allow low-risk read-only lookups and safe refactors.
Log fields
timestampagent nameenvironmentsession idtrace idtoolactionstatusrisk scorerisk levelpermission decisionapproval statusreasoning summaryrollback referenceredacted metadata
{
"rules": {
"hardcoded_secrets": process.env.HARDCODED_SECRETS,
"sql_injection": "block",
"missing_rate_limit": "warn",
"auth_changes": "require_approval",
"destructive_shell": "block"
}
}{
"agent": "coding-agent",
"tool": "file",
"action": "write",
"status": "gated",
"risk_level": "high",
"decision": "require_approval",
"metadata": "[redacted context]"
}allow
Continue immediately.
warn
Surface risk for review.
block
Stop the action or code change.
require_approval
Pause for human review.
log
Record searchable evidence.
09 / Integrations
VibeLint fits beside the tools that already run your agents.
It is not an agent builder. It is the security, permission, and evidence layer around IDE agents, workflow builders, agent frameworks, and custom runtimes.
IDE & MCP agents
- Codex: Guided setup preserves unrelated MCP servers and updates only mcp_servers.vibelint.
- Cursor: Guided setup preserves unrelated MCP servers and updates only mcpServers.vibelint.
- Claude Code: Guided setup preserves unrelated MCP servers and updates only mcpServers.vibelint.
- Antigravity: Guided setup preserves unrelated MCP servers and updates only mcpServers.vibelint.
- Devin Desktop (formerly Windsurf): Guided setup preserves unrelated MCP servers and updates only mcpServers.vibelint.
Workflow builders & frameworks
- n8n: Use HTTP nodes to call VibeLint permission checks and log agent actions.
- Zapier Agents: Call VibeLint API from Zapier actions for permission and logging.
- Make: Use HTTP modules to connect AI agent workflows to VibeLint.
- Dify: Add VibeLint API calls as tool nodes in Dify workflows.
- Flowise: Connect Flowise agents to VibeLint via HTTP tool nodes.
- LangChain / LangGraph: Add VibeLint permission checks as tool calls in LangGraph agents.
- CrewAI: Wrap VibeLint API calls as CrewAI tools for agent security.
- OpenAI Agents SDK: Register VibeLint as a function tool in the OpenAI Agents SDK.
- Custom agents: Any runtime that can call HTTP can use VibeLint for permissions, logs, and approvals.
Custom agent runtimes
- Use structured HTTP calls for action logs and permission checks.
- Use agent identities to separate bots, environments, and workflows.
- Keep real agent keys in the dashboard and use placeholders anywhere public.
- Any language or framework that can call HTTP works with VibeLint.
POST https://vibelint.dev/api/agent-actions/log
Authorization: Bearer vl_agent_your_key
{
"tool": "stripe",
"action": "issue_refund",
"status": "executed",
"reasoning_summary": "Customer was double charged",
"rollback_available": true
}POST https://vibelint.dev/api/agent-permissions/check
Authorization: Bearer vl_agent_your_key
{
"tool": "database",
"action": "update_records",
"input": { "record_count": 12 },
"rollback_available": false
}10 / Plans
Choose the plan that matches how seriously you use AI agents.
Start with Free to try local AI code protection. Upgrade to Pro when AI is part of your daily development workflow and you want the full detector suite, action logs, permissions, approvals, and dashboard history. Use Max when you need higher production-volume limits.
Free
$0 forever
Best for: Trying VibeLint and protecting personal AI coding workflows.
Includes: Unlimited local MCP protected checks, 500 dashboard-synced protected checks per month, Free detector set, pre-commit hook, remediation hints, 1 project, 1 agent, 100 monthly action logs, and 7-day dashboard retention.
Pro
$19.99/month, $199/year, or limited-time $199 lifetime Pro
Best for: Builders using AI coding tools daily and needing full detector coverage plus visibility.
Includes: Unlimited local MCP protected checks, 1,000 dashboard-synced protected checks per month, all 11 detectors, prompt injection and LLM output execution checks, 10 projects, 10 agents, 25,000 monthly action logs, policies, approvals, security score, weekly digest, and 30-day retention.
Max
$49/month or $469/year
Best for: High-volume production agent security.
Includes: Unlimited local MCP protected checks, 10,000 dashboard-synced protected checks per month, everything else in Pro, 50 projects, 50 agents, 250,000 monthly action logs, 250 policies, 2,500 rules, 2,500 pending approvals, and 90-day retention.
Enterprise
Contact VibeLint
Best for: Security review, procurement, custom terms, and regulated team needs.
Includes: Custom terms, security review, procurement-friendly billing, priority engineering support, and implementation guidance.
11 / Trust and privacy
Security visibility without sending more data than needed.
VibeLint is designed to give you useful security signals and agent audit history while keeping the scanning workflow lightweight. The scanner runs locally by default, and dashboard sync focuses on findings, summaries, and action evidence instead of full source files.
Local code checks
The scanner runs locally by default so VibeLint can review generated code without making source-code upload the core workflow.
Dashboard findings
When scan history syncs to the dashboard, it is focused on security metadata such as detector type, severity, project/file context, descriptions, fix hints, and timestamps.
Agent audit context
For HTTP agent logs and permission checks, VibeLint stores the structured context your agent sends and masks common sensitive key names before showing it in the audit trail.
Security contact
If you find a vulnerability, email hi@vibelint.dev with the affected area, reproduction steps, impact, and enough detail for the issue to be confirmed.
12 / FAQ
Common questions before adopting VibeLint.
Short answers for the things people usually want to know before they install VibeLint, connect an agent, or choose a paid plan.
Is VibeLint a code scanner?
Yes, but it is more than that. VibeLint is an AI agent security layer that also includes agent action logs, permission policies, approval workflows, and audit history for agent behavior.
Is VibeLint only for Cursor?
No. Guided MCP setup supports Cursor, Claude Code, Antigravity, Codex, and Devin Desktop (formerly Windsurf) on Windows and Mac/Linux. Workflow builders and custom agent frameworks can use the HTTP agent APIs.
How do I connect my AI agent to VibeLint?
For IDE agents, open the dashboard and run the one guided setup command for your editor and OS. It installs a private runtime, authorizes your device in the browser, and configures only the VibeLint MCP entry. For workflow builders and agent frameworks, use the VibeLint HTTP API with your agent key. Call the permission check endpoint before sensitive actions and the action log endpoint after each action.
Does VibeLint replace security reviews?
No. VibeLint catches common AI-generated risks early and gives reviewers better evidence, but teams should still use normal security review practices.
Does VibeLint store my source code?
Code scanning runs locally by default. Dashboard sync focuses on metadata, summaries, detector names, severity, file context, decisions, approvals, and redacted action context rather than raw source files.
Can VibeLint block an AI agent?
Yes, VibeLint can return block decisions for permission checks and flag or block risky code depending on workflow configuration and policy.
What languages does VibeLint support?
The current public examples emphasize JavaScript, TypeScript, and Python-style workflows. Detector behavior depends on the code, scanner package, and rule coverage available to your plan.
Is VibeLint worth paying for?
If you only want to try local AI code checks, start with Free. If AI is now part of your daily development workflow, Pro is usually the useful tier because it adds the full detector suite, action logs, permission policies, approvals, security score, and longer history.
Which plan should I start with?
Start with Free if you are evaluating the scanner. Choose Pro if you want serious daily coverage for AI coding and agent workflows. Choose Max if you run many agents or need higher scan, log, policy, rule, and retention limits.
How is VibeLint different from Snyk, Semgrep, or GitHub Advanced Security?
Those tools are valuable, but they usually focus on repository, dependency, or pull request security. VibeLint focuses on AI agent workflows, pre-write checks, permissions, and action evidence.