Everything Enclave does
A plain list of what Enclave's security agents do today, area by area.
Code security analysis
Repository scans
Enclave maps each repository's components and attack surfaces, then hunts for vulnerabilities in them. Later scans re-review only the surfaces whose code changed.
Verified findings only
An analyst agent works each attack surface and a separate verifier traces the code flow and rates severity. Only verified findings are shown, and duplicates are removed within a scan and against your existing findings.
Exploitability in your environment
Each finding is judged against the graph of your code and cloud, so severity reflects whether the weakness is actually exposed.
Pull request security review
Review on every pull request
When someone opens or pushes to a pull request on GitHub or GitLab, including self-managed GitLab, Enclave reviews the change and comments on the exact lines involved.
What it looks for
Authentication, access control, secrets, cryptography, data exposure, configuration, and injection flaws such as SQL injection, XSS, and SSRF.
A status check you can gate on
The Enclave / PR analysis check fails on high or critical concerns. Severity accounts for how exposed the affected service is in your cloud.
Discuss and fix in the thread
Reply to a comment to challenge it, resolve the thread to dismiss it, or write @enclave review to run the review again. Each finding comes with a ready prompt for your coding assistant.
Software composition analysis (SCA)
Dependency inventory
Enclave lists every pinned package across your repositories and highlights vulnerable versions. It reads npm, pnpm, Yarn, and Bun; Python requirements, Poetry, uv, PDM, and Pipenv; Go modules; Cargo; RubyGems; Composer; Maven and Gradle; and .NET projects.
Advisories that stay current
Advisories come from OSV. When one is published, edited, or withdrawn, your inventory reflects it within about 10 minutes, with no rescan.
Malicious package checks on pull requests
When a pull request adds or updates an npm or PyPI package, agents inspect its install scripts, maintainer history, download counts, registry metadata, and the diff between versions, and comment on supply-chain concerns.
AI penetration testing
Attack scenarios you approve
Enclave crawls your application in an isolated MicroVM and proposes attack scenarios. It attacks only the scenarios a person approves.
Exploits confirmed twice
A scenario becomes a finding only when a second, independent pass confirms the exploit.
Authenticated testing
Pentests sign in as the test identities you save, including accounts protected by TOTP and email codes.
Traffic limited to your scope
Every run sends its traffic through its own proxy, limited to the exact hosts in scope.
Proof of blind vulnerabilities
Enclave catches DNS, HTTP, HTTPS, and TCP callbacks to prove vulnerabilities that give no direct response.
Retesting
A retest that no longer reproduces a confirmed exploit marks its finding resolved. If the exploit works again, the finding reopens.
Cloud discovery and the security graph
Connected clouds
Connect AWS, Google Cloud, Azure, Vercel, Cloudflare, and Kubernetes. Enclave reads each one with short-lived, read-only credentials it never stores.
One graph of your environment
Discovery runs when you connect a cloud and again every night, mapping services, DNS, and ingress routing into the same graph as your repositories and applications.
Private infrastructure
Reach hosts that are not on the public internet through SSH targets and Tailscale.
Threat feed and CVE research
A public threat feed
Agents research new CVEs and advisories as they land and publish a report on each: how exploitable it is and what an attacker needs.
Am I exposed?
From any CVE report, one click opens a chat that checks your own repositories and cloud for exposure.
Security chat
Ask about your environment
Ask about findings, code, cloud assets, and the security graph. The chat can run read-only cloud commands and work with files you upload.
Desktop app
Local chats in the desktop app work on your own machine, and ask for your approval before each command or file edit.
Slack
Chat with Enclave in a Slack thread, approve agent requests there, and receive new findings in a channel or by direct message.
Skills and subagents
Teach Enclave reusable skills, and let a chat hand parts of an investigation to subagents.
Workflows and custom agents
Workflows that start themselves
A workflow can run on a schedule or when a finding is created or verified, a pull request is updated, or a repository is pushed or connected.
Custom agents
Define your own agents as roles in a workflow, each with its own instructions and tools. A role you give repository write access can push a branch and open a pull request.
Marketplace
Start from workflows that others have published, then adapt them to your team.
Control over every run
Pause or stop a run, set a budget for it, and inspect its records and outputs.
Integrations
Source control
GitHub, GitLab including self-managed, and Azure DevOps.
Issue trackers
Workflows file issues in Jira and Linear, and findings link to Azure Boards.
MCP server
Add Enclave to Claude, Cursor, or another MCP client. Your assistant can read and triage findings, read a pull request's scan, read the cloud graph, work with workflows, and read the threat feed, always with your own permissions.
MCP connections
Connect outside MCP servers so Enclave's agents can use your other tools.
Notifications and exports
Send findings to Slack or email, and export any finding as PDF or Markdown.
Administration and access
Roles
Invite people as owner, admin, member, or a custom role, each limited to the scopes you grant.
Audit log
A record of what people and agents did in your organization.
Your own model API keys
Add your organization's own Anthropic or OpenAI key, and that provider's calls run on your account.
Sign-in
Sign in with Google, Microsoft, or an email code.