BYOAI code auditing: security + bugs

Find security flaws and software bugs with the AI agents you already use.

Run security audits or bug hunts manually, or use our support-assisted GitHub integration for incremental pull-request work. TheSpider runs Claude Code, Codex, OpenCode, and other commands on your machine, then keeps every finding deduplicated, stored, and ready to triage in one place.

$ curl -fsS thespider.xyz/healthz → ok  ·  bring your own models · keep your audit history · hosted SaaS, local execution

thespider-agent — zsh
$ thespider-agent run --once --max-slices 1
●  project acme/payments-api (profile: bugs)
→  claim slice #142 … ok  (auth/, api/handlers/)
→  render prompt … 8.2k tokens
→  run claude … done in 3m12s (checkout not uploaded to TheSpider)
  3 findings reported · 1 duplicate suppressed by AI adjudication
   HIGH  concurrent refresh can overwrite a newer session
   MED   pagination drops records after a deleted cursor
   LOW   retry loop ignores the configured backoff limit
$ 

// two audit modes

Two ways to hunt. One findings history.

Security audits

Find vulnerabilities, broken authorization, unsafe trust boundaries, injection risks, leaked secrets, and other exploitable weaknesses.

Bug hunts

Find correctness defects, broken edge cases, race conditions, state inconsistencies, error-handling failures, and regressions.

Every project contains both profiles with independent methodologies and campaigns. Findings, dedupe, triage, fixes, and status remain shared project-wide.

// why thespider

More than a one-off AI report.

TheSpider is the system of record for AI-assisted code audits. It remembers what has already been found, combines duplicate reports across models and runs, and gives your team one queue for triage, remediation, and retesting.

Your checkout stays off our server

A thin Rust agent runs model commands on the machine that has the checkout. TheSpider receives git metadata and model output, not your source tree. Your chosen model CLI may still send context to its own provider.

Bring your own AI

Use Claude Code, Codex, OpenCode, Gemini, aider, or any compatible local command. Mix models in one project to benefit from their different strengths without locking your audit history to one provider.

Persistent findings

Findings survive individual runs. TheSpider tracks identity, status, evidence, and sightings over time, then collapses duplicate reports across models and branches into one finding with full history.

Run a repeatable methodology

TheSpider builds campaigns from your configured passes, buckets, and enabled models, then tracks each work unit from claim through result ingestion.

Triage and gate

Review and update findings in the dashboard. Run full audits from the agent; the currently support-assisted GitHub integration creates incremental campaigns that CI can gate.

A hosted brain, without a hosted checkout

The SaaS service owns methodology, dedupe, findings, and team workflow. Only the prebuilt agent and your chosen model CLIs run on your Linux machine or WSL environment.

// how it works

From signup to first audit

01

Install the verified client

The first-party installer selects and verifies the matching Linux binary without a source checkout.

02

Run setup in the repository

Setup opens browser authorization, safely selects or creates the matching project, and binds your agent (Claude Code, Codex, or OpenCode) to Security or Bugs.

03

Approve one bounded call

Review the full planned matrix, then claim exactly one slice. Remaining work stays running and visible.

04

Triage in the Cloud

Reports are parsed, deduplicated, and added to one shared findings history for remediation and retesting.

thespider-agent (Rust)   HTTPS / bearer      brain server (Bun)
• enroll + token      ── claim/results ──▶ • orgs · RBAC · quotas
• model → command     ◀── prompts ───────  • parse · dedupe · triage
• git detect + runs                        • web UI + PR gate
    runs in your checkout                    does not receive your tree

Leaves your machine

  • Git metadata & changed paths
  • Raw model stdout, including findings and cited evidence
  • A short stderr tail for diagnostics

TheSpider does not upload

  • Your source tree
  • Your model credentials & CLI auth

Parsed findings are scrubbed before report rows are written. Raw stdout and stderr artifacts are stored as received unless you enable client-side --redact.

// pricing

Simple org plans. No custom quotes.

You bring the models. We sell the audit brain. Self-serve checkout — the listed price is the price.

Free
$0
  • 1 project
  • 200 slice claims / period
  • 3 seats
  • Fingerprint dedupe
Create account
Pro
$49/mo
  • 10 projects
  • 10,000 slice claims / period
  • 10 seats
  • AI-assisted dedupe
Start on Free, upgrade in settings
Business
$149/mo

or $1,490/year

  • 50 projects
  • 50,000 slice claims / period
  • 50 seats
  • AI-assisted dedupe
  • OIDC SSO
Start on Free, upgrade in settings

On-prem is Business billed yearly: $1,490/year, same 50 seats / 50 projects / SSO. Docs-only support. Compare deployment options

// evaluating deployment options?

TheSpider is also available on-prem.

The on-prem edition is the Business plan as a single-tenant deployment — $1,490/year, same seats, projects, and SSO — for organizations with data-residency, private-network, or air-gap requirements. It supports Docker Compose or Kubernetes, customer-controlled Postgres and object storage, and local or internal model endpoints. Public setup instructions on this site are for TheSpider Cloud.

Compare deployment options

// get started

Start in your coding agent.

Install the hosted skill and paste a short prompt, or fetch the one-shot setup wizard. Both paths install the verified client, run browser-authorized setup, and ask before one model call.

CLI + skill

Prefer the local agent

Install TheSpider's skill into Claude Code, Codex, OpenCode, and the other agents npx skills detects. Then paste the prompt.

  1. 1. Install the skill

    npx skills add https://thespider.xyz -g
  2. 2. Paste this prompt

    Get me started with TheSpider.

URL wizard

Let an agent fetch the setup script

No skill install. Paste this into an agent that can fetch a URL — it follows startup.md.

Get me started with TheSpider: https://thespider.xyz/startup.md

Linux or WSL · no server clone · no Docker, Bun, or Rust toolchain

Put your AI agents to work finding bugs.

Start with a security audit or general bug hunt. Run one model or many, keep your checkout off TheSpider's server, and build a findings history that improves with every campaign.