Agent-powered vulnerability scanning for large-scale codebases — on your own infrastructure.

Quick start · Documentation · Models & agents · Configuration

---

^{Daniel Bannert's open source work is supported by the community on GitHub Sponsors}

---

What is deepsec?

deepsec is an agent-powered vulnerability scanner that you run on your own infrastructure, optimized for on-demand review of all code in existing large-scale repos. It is designed to surface hard-to-find issues that have been lurking in applications for a long time.

It is configured to use the best models at maximum thinking levels (tunable via --thinking-level, see models), meaning scans can cost thousands or even tens-of-thousands of dollars for large codebases. Our customers have found the cost worth it for how quickly they were able to patch vulnerabilities that would have otherwise gone unfixed.

For large codebases, work fans out across worker machines in parallel. If a run is interrupted or errors out partway through, just re-run the same command — deepsec picks up where it left off, skipping files it already analyzed and only investigating the rest.

What's improved vs. the original

This fork builds on vercel-labs/deepsec and adds:

• OpenCode as an agent backend (--agent opencode). Both OpenCode runtime generations work — v1 (npm opencode-ai 1.x, bundled for sandbox workers) and the new OpenCode v2 (2.0.x), detected automatically at startup. The v2 runtime cannot be driven by the published SDK (a different API surface under /api/…, mandatory basic auth, an async prompt/wait inbox), so the backend speaks both protocols itself. Every batch runs in a private, password-secured OpenCode server with a read-only agent — only read, glob, grep, and list; shell, edits, network tools, subagents, external directories, LSP, and skills are denied. Works for process, revalidate, and the one-shot init repository analysis.
• Vercel AI Gateway on the OpenCode backend. On the v2 runtime, the native vercel/ provider takes one AI_GATEWAY_API_KEY for every model behind the gateway (--model vercel/anthropic/claude-opus-4-8). The standard gateway credential expansion also keeps working — it is translated into v2 provider baseURL overlays, with the bearer token bridged to the x-api-key form the v2 anthropic provider reads.
• No Vercel account needed for your own key. init --model-auth direct no longer prompts for a Vercel login (upstream issue #164): the platform link — which only enables the optional Vercel Sandbox distribution — is reused silently when Vercel credentials exist and skipped otherwise, while your provider key is still resolved and verified.
• Config matcher filtering works. matchers: { only, exclude } in deepsec.config.ts is honored by scan (upstream issue #36); unknown slugs in the config fail loud instead of being silently ignored.
• Your Codex subscription is respected. A codex login subscription wins over a stray OPENAI_API_KEY in the environment — matching the codex CLI's own precedence — unless explicit gateway/base-URL routing is configured (upstream issue #32).
• Nine upstream fix MRs merged that repair open issues left sitting upstream: severity ordering (#48/#61), unknown --matchers slugs (#34/#75), triage data in exports (#64/#71), triage verdict matching (#118/#119), unparseable triage output (#120/#121), token-metrics table layout (#135/#149), non-coherent host mounts (#159/#160), provider policy refusals parsed as findings (#92/#137), and macOS test flakiness (#51/#50). Upstream issues #30, #29, #33, and #91 were verified as already fixed in current main.
• Published as @anolilab/deepsec with the same deepsec CLI and release automation via npm trusted publishing — no tokens in CI.

Quick start

From the root of the repository you want to scan:

npx @anolilab/deepsec init

The command guides you through everything. It asks you to pick an AI model (with benchmark scores and prices to compare) and how to pay for model usage — your own OpenAI/Anthropic API key, or Vercel AI Gateway — and then works unattended: it studies your codebase, scans it, and runs the AI review. The only thing it adds to your repository is a .deepsec/ folder where all of its state and findings live.

If the run is interrupted for any reason — Ctrl-C, lost connection, a spending limit — run npx @anolilab/deepsec init again and it continues where it left off. To cap what a run may spend or how long it may take:

npx @anolilab/deepsec init --max-cost-usd 100 --max-duration 2h

When the scan finishes, get a readable report:

cd .deepsec
pnpm deepsec export --format md-dir --out ./findings

For later scans, work from inside .deepsec/:

pnpm deepsec scan        # fast pattern scan, free
pnpm deepsec process     # AI review of new candidates
pnpm deepsec revalidate  # optional, cuts false-positive rate
pnpm deepsec export --format md-dir --out ./findings

The getting started guide covers all of this in more detail, including using your own OpenAI or Anthropic API key and running from CI or a coding agent.

Docs

After initialization, agents can read the exact documentation matching the installed CLI at .deepsec/node_modules/@anolilab/deepsec/SKILL.md and .deepsec/node_modules/@anolilab/deepsec/dist/docs/. Setup errors expose these as absolute machine-readable paths.

• Getting started — set up and run your first scan
• Reviewing changes — process --diff and CI gating
• Supported technology — built-in coverage
• Generated and hand-authored matchers
• Configuration
• Plugins
• Models — agent backends, thinking levels, credentials, recommended models & ensemble strategy
• Project link and credentials
• Architecture
• Data layout
• FAQ
• Samples
• Contributing

AI agents

The AI review runs through an interchangeable agent backend, selected with --agent (or defaultAgent in deepsec.config.ts). Same prompt, same JSON output contract — you can mix backends within a repo and compare models under the same workload.

Backend	SDK	Default model
codex (default)	@openai/codex-sdk	gpt-5.5
claude	@anthropic-ai/claude-agent-sdk	claude-opus-4-8
opencode	@opencode-ai/sdk/v2 + opencode runtime	anthropic/claude-opus-4-8
pi	@earendil-works/pi-coding-agent	zai/glm-5.2

pnpm deepsec process --project-id my-app --agent claude
pnpm deepsec process --project-id my-app --agent opencode --model anthropic/claude-opus-4-8

OpenCode backend

The opencode backend speaks both OpenCode runtime generations — the v1 runtime (npm opencode-ai 1.x, also bundled for sandbox workers) and the new OpenCode v2 (2.0.x) — detected automatically at startup, so it works with whatever opencode you have installed. Each batch runs in a private, password-secured OpenCode server with a read-only agent: only read, glob, grep, and list are allowed; shell, edits, network tools, subagents, external directories, LSP, and skills are denied.

For a local run without environment credentials, connect a provider first (run opencode, then use /connect) and use its provider/model id:

pnpm deepsec process --project-id my-app \
  --agent opencode \
  --model openai/gpt-5.5

Through the Vercel AI Gateway, the v2 runtime also accepts the native vercel/ provider — one AI_GATEWAY_API_KEY covers every model behind the gateway:

AI_GATEWAY_API_KEY=vck_… pnpm deepsec process --project-id my-app \
  --agent opencode \
  --model vercel/anthropic/claude-opus-4-8

See models for the full backend reference, thinking levels, credential routing, and the recommended models & ensemble strategy for maximum issue discovery.

AI provider

By default, deepsec routes model calls through Vercel AI Gateway, which gives access to every major model without provider-specific keys — one AI_GATEWAY_API_KEY covers every agent backend above. You can instead bring your own key — OpenAI, Anthropic, or a custom HTTPS provider — by passing --model-auth direct with --ai-provider and --ai-api-key-env to init; no Vercel account is needed in that mode. Deepsec only ever stores the name of the environment variable holding your key, never the key itself. See project link and credentials for the full reference.

When running locally, deepsec can also reuse existing Claude, Codex, Pi, or OpenCode provider authentication — including a local opencode CLI login (opencode → /connect) — for evaluation-scale scans.

If a process or revalidate run halts because the upstream credential ran out of quota or credits, deepsec stops gracefully and tells you where to top up. Re-run the same command afterward and it picks up where it left off.

Distributed execution (optional)

Large monorepos can fan work across Vercel Sandbox microVMs:

pnpm deepsec sandbox process --project-id my-app --sandboxes 10 --concurrency 4

Setup already verified the Vercel connection, so this needs no extra onboarding. The local working tree is tarballed and uploaded; .git is excluded. Model credentials remain host-side and are injected only at the selected egress host.

Security model of deepsec itself

Treat deepsec like a coding agent with full shell access on the enviroment that it is running on. It is designed to run on trusted inputs (your source code) but you may still be concerned about prompt injection due to external dependencies or vendored code.

Running on a sandbox (see above) does limit the potential exposure substantially:

• The API keys for the coding agents are injected outside of the sandbox and hence cannot be exfiltrated
• For the worker sandboxes, network egress from the sandbox is limited to coding agent hosts (Egress is allowed during the bootstrap process, but this does not run the coding agent)

Workflow reference

Command	What it does
scan	Find candidate sites with regex matchers (fast, no AI)
process	AI investigation; emits findings + recommendation
process --diff	PR-mode: scan + investigate only files changed in a diff
triage	Lightweight P0/P1/P2 classification (cheaper model)
revalidate	Re-check existing findings; checks git history for fixes
enrich	Add git committer info + (with a plugin) ownership data
report	Markdown + JSON summary for one project
export	Per-finding JSON or directory of markdown files
metrics	Cross-project counts: severities, vulns by type, TPs
status	Snapshot of the project mirror
sandbox <cmd>	Run any of the above on Vercel Sandbox microVMs

Contributing

Start with CONTRIBUTING.md. For security reports, see SECURITY.md. For community guidelines, see CODE_OF_CONDUCT.md.

License

Apache 2.0. See LICENSE and NOTICE.