# Enclave > A fleet of autonomous security agents that finishes scoped security jobs end to end for AI-native companies. Both sides have agents now. Attackers run theirs without guardrails or approval chains. Most security agents just generate more work: scan, flag, hand you a list. Enclave's agents carry the work through instead: they verify reachability and exploitability against your live environment, write each fix as a PR or infrastructure-as-code change, route it to the owner, and leave every merge to your engineers for approval. After deploy, they re-run the original attack path and confirm it no longer works. Enclave connects repositories and cloud inventory into a living security graph. Agents run against it, reasoning about reachability, identity, exposure, and attacker preconditions, then drive each confirmed finding to a fix your team approves. The longer it runs, the more it knows about your architecture, and the faster it gets. ## Platform Capabilities Not a pipeline of steps: a fleet of agents working one live model of your environment, in parallel and continuously, from first scan to verified closure. - **Live security graph**: Services, repos, cloud accounts, and existing scanners connected into a single graph of components, entry points, trust boundaries, and ownership. Every pull request and infrastructure change updates it. Covers AWS, GCP, and Azure. - **Agentic investigations**: Agents run code review, SAST, DAST, cloud misconfiguration, secrets, and supply chain analysis, alongside CVEs, pre-CVE advisories, and threat intel, reasoning about reachability, identity, exposure, and attacker preconditions the way a security researcher would. - **Verification and triage**: Every finding is tested against the graph for reachability and exploitability in the live environment, then ranked by KEV status and business impact. What reaches the team is confirmed, prioritized exposure, not a CVSS-sorted list. - **Batched remediation**: Agents write each fix as a PR or infrastructure-as-code change, dispatched to the coding agents you already license (BYOK), and route it to the owner in a plan optimized to close the most exposure with the least effort. Nothing merges without a human. - **Closure verification**: After deploy, agents re-run the original attack path. If it still works, the process restarts; if not, the evidence is logged: exploit path, exact change, post-deploy re-test. An audit trail for every closed finding. - **Security chat**: Query findings, attack paths, affected assets, ownership, and prioritization reasoning: the same live context the agents work from. Bring your own model API keys (Anthropic, OpenAI, or Google) or run models through Enclave; the harness is purpose-built for security research. Approval gates are set per repository. Deploys as SaaS, in your private VPC (AWS, GCP, or Azure), or fully air-gapped for regulated and classified environments. Built for CISOs, VP Security, and CloudSec and AppSec security engineers. Harnesses the stack you have: SAST/SCA, CSPM, vulnerability scanners, SIEM/SOAR, ticketing systems, GRC/risk registers, pen-test programs, and manual code review. ## Features The full list, area by area, is at [enclave.ai/platform](https://enclave.ai/platform). - **Code security analysis**: Repository scans map each repository's components and attack surfaces, then agents hunt for vulnerabilities in them. A separate verifier traces the code flow and rates severity, and only verified findings are shown. Later scans re-review only what changed. - **Pull request security review**: On GitHub and GitLab (including self-managed), Enclave reviews pull requests in the repositories where you turn reviews on and comments on the exact lines, covering authentication, access control, secrets, cryptography, data exposure, configuration, and injection flaws such as SQL injection, XSS, and SSRF. An `Enclave / PR analysis` status check fails on high or critical concerns. Reply to a comment to challenge it, or write `@enclave review` to run it again. - **Software composition analysis (SCA)**: A dependency inventory across npm, pnpm, Yarn, Bun, Python (requirements, Poetry, uv, PDM, Pipenv), Go, Rust (Cargo), Ruby, PHP (Composer), Java (Maven, Gradle), and .NET, matched against OSV advisories that update within about 10 minutes, with the version that fixes each one. See [Dependency protection](https://enclave.ai/solutions/dependency-protection). - **Malicious package detection**: When a pull request adds or updates an npm or PyPI package, agents check known vulnerabilities, install scripts, maintainer history, download counts, and the code changed between versions, and flag obfuscated code or data sent to unknown hosts. - **AI penetration testing**: Enclave crawls your application in an isolated MicroVM, proposes attack scenarios, and attacks only those a person approves. A second, independent pass confirms each exploit. Tests sign in as saved test identities (including TOTP and email codes), send traffic only to in-scope hosts, catch DNS, HTTP, HTTPS, and TCP callbacks to prove blind bugs, and retest confirmed findings. - **Cloud discovery**: Read-only connections to AWS, Google Cloud, Azure, Vercel, and Cloudflare, plus an in-cluster agent for Kubernetes, mapped nightly into the same graph as your repositories, plus SSH targets and Tailscale for private hosts. - **Threat feed and CVE research**: Agents research critical and actively exploited CVEs and publish a report on each. "Am I exposed?" checks your own repositories and cloud. - **Security chat**: Ask about findings, code, cloud assets, and the graph. Also in the desktop app (with approval for every command and file edit) and in Slack threads. - **Workflows and custom agents**: Workflows run on a schedule or on events (finding created or verified, pull request updated, repository pushed or connected). Define custom agents as workflow roles, start from the Marketplace, and set a budget for each run. - **Integrations**: GitHub, GitLab, and Azure DevOps; Jira and Linear; Slack; an MCP server so Claude, Cursor, and other MCP clients can read and triage findings with your permissions; and outbound MCP connections. - **Administration**: Owner, admin, member, and custom roles; an audit log; sign-in with Google, Microsoft, or an email code; PDF and Markdown export of findings. ## Pricing - **Pay as you go**: $0 with a card on file. 10 PR reviews per month, PRs from agents included, then paid from credits; 1 free repo scan, then paid from credits; free dependency scans and CVE alerts; Marketplace workflows paid from credits. Custom workflows, chat and pentests are locked. Unlimited app members, cloud deployment. - **Pro**: $50 per seat per month, 10 to 50 seats; a seat is a person whose PRs Enclave reviews, app member or not. Unlimited PR reviews for people with a seat, with a fair use of 150 per seat per month; PRs from agents need no seat and are paid from credits. 1 repo scan per seat per month (shared, then paid from credits), free dependency scans and CVE alerts, $10 in credits per seat each month, chat and custom workflows, 3 pentests per month (then paid from credits), unlimited app members, cloud deployment. - **Enterprise**: annual contract with negotiated seats. Unlimited PR reviews, repo scans and pentests at contract volume, Marketplace and custom workflows, chat included or BYOK, unlimited app members with SAML and SCIM, BYOC, BYOK or on-prem deployment. - **Startups**: $1,000 in usage credits for VC-funded startups from Seed to Series A with less than $2M in revenue over the past 12 months. - **Open source**: contributed Enclave credits for maintainers of widely used open-source projects. See [enclave.ai/pricing](https://enclave.ai/pricing). ## Get a Demo Get a demo at [enclave.ai/demo](https://enclave.ai/demo) to see agents carry a finding from scan to verified closure. ## Solutions - [Remediation](https://enclave.ai/solutions/remediation): Automated vulnerability remediation with fix pull requests, retested after deploy. - [Continuous penetration testing](https://enclave.ai/solutions/continuous-pentest): AI pentesting; autonomous, continuous testing of live attack paths. - [Finding triage & verification](https://enclave.ai/solutions/triage): Reachability and exploitability analysis that separates verified risk from noise. - [Code security review](https://enclave.ai/solutions/code-security): Find security weaknesses in code context. - [Pull request security review](https://enclave.ai/solutions/pr-review): Security feedback in the review workflow. - [Dependency protection](https://enclave.ai/solutions/dependency-protection): Investigate package changes before they ship. - [CVE monitoring](https://enclave.ai/solutions/cve-monitoring): Check new advisories against your environment. - [Custom security agents](https://enclave.ai/solutions/custom-agents): Agents built for your security workflows. ## Comparisons - [Enclave vs Snyk](https://enclave.ai/compare/enclave-vs-snyk): Agentic application security compared with Snyk's developer security scanners. - [Enclave vs Wiz](https://enclave.ai/compare/enclave-vs-wiz): Agentic application security compared with Wiz's cloud security platform. - [Enclave vs Aikido](https://enclave.ai/compare/enclave-vs-aikido): Agentic application security compared with Aikido's all-in-one security suite. - [Enclave vs XBOW](https://enclave.ai/compare/enclave-vs-xbow): Agentic application security compared with XBOW's autonomous pentesting. ## Programs - [Enclave for Open Source](https://enclave.ai/open-source): Credits for code auditing and PR security review of open-source projects, with findings shared privately with maintainers. - [Startups program](https://enclave.ai/startups): VC-funded startups from Seed to Series A get $1,000 in Enclave usage credits. - [Research Fellowship](https://enclave.ai/research-fellowship): $1,000 every month for independent security researchers to use for research on Enclave, with findings disclosed and published with credit. ## Research and Company - [Research](https://enclave.ai/research): Original security research from Enclave, including the AI Hacking Race benchmark and published findings. - [AI Hacking Race](https://enclave.ai/hackingrace): Leaderboard comparing verified runs, time, and cost across AI models. - [Events](https://enclave.ai/events): Webinars, dinners, and conferences with Enclave's research team, and past sessions on demand. - [About](https://enclave.ai/about) - [Careers](https://enclave.ai/careers) - [Contact](https://enclave.ai/contact) - [Threat Feed](https://app.enclave.ai/threat-feed): Agent research on new CVEs and pre-CVE advisories. - [Docs](https://docs.enclave.ai) ## Links - [Homepage](https://enclave.ai/) - [Platform features](https://enclave.ai/platform): Every Enclave feature by area, including SCA, pull request review, AI penetration testing, cloud discovery, MCP server, and admin controls. - [Agentic AppSec](https://enclave.ai/agentic-application-security) - [Agentic application security vs. traditional AppSec](https://enclave.ai/agentic-application-security-vs-traditional-appsec) - [Security code review](https://enclave.ai/security-code-review) - [Self-hosted deployment](https://enclave.ai/self-hosted): Run Enclave in your own VPC or air-gapped, with your own Anthropic, OpenAI, or Google model API keys. - [Pricing](https://enclave.ai/pricing) - [Download the desktop app](https://enclave.ai/download) - [Blog](https://enclave.ai/blog) - [Get a Demo](https://enclave.ai/demo) - [Trust Center](https://trust.enclave.ai) - [Twitter/X](https://x.com/EnclaveAI) - [LinkedIn](https://www.linkedin.com/company/enclaveai) ## Optional - [Enclave AI reference](https://llms.enclave.ai/llms.txt): Index of Enclave's reference pages for AI agents and its public question endpoint, with the supported request formats.