An exploit proves the risk is real. Closing it is the rest of the job, and that's the part Keygraph runs: fixes land as reviewable pull requests, the original exploit re-runs to verify them, and the finding is tracked to verified closure.
XBOW is built to do one thing extraordinarily well: autonomously identify what is exploitable, validate it by exploitation, and deliver reproducible proofs-of-concept and compliance-ready reports.
But a penetration test assesses only the targets it is pointed at. Everything around the exploit, including the rest of the attack surface and the work of fixing a finding and tracking it to closure, sits outside its scope. This is where Keygraph takes a broader approach. The Keygraph platform runs the same offensive testing, black-box or white-box, then adds continuous SAST, SCA, and secrets scanning across the codebase, correlates every finding to the source that produced it, and re-runs the original exploit after the fix ships to confirm the vulnerability is closed. All of it can deploy self-hosted in your own environment, embedded in your SDLC, with one system of record from vulnerability identification through verified remediation.
XBOW set out to automate black-box penetration testing. A penetration test, however strong, is one assessment technique within a broader security assessment program: your team still has to cover the attack surface the test never reached, remediate what it finds, and re-test to verify the fixes. Keygraph pairs the same autonomous offensive testing with the platform that runs the rest: a force multiplier for security teams and pentesters, not a replacement.
XBOW
autonomous offense
A formidable autonomous attacker, built for exploitation.
Keygraph
the AppSec platform
Exploitation is one capability, not the whole job.
XBOW and Keygraph both do the offensive half: autonomous vulnerability identification, real exploitation, and validation with reproducible proof. The difference is everything around and after the exploit: who assesses the rest of the attack surface, who drives remediation, and who re-tests to prove the fix held.
XBOW identifies and validates vulnerabilities against the running app and can re-test a fix on demand, but writing and tracking the fix is left to your team. Keygraph continues through Fix and Verify: remediation lands as a reviewable pull request, and the original exploit is re-run to confirm the fix.
Both discover vulnerabilities autonomously. XBOW maps the running app; Keygraph also sweeps the code behind it: SAST, SCA, and secrets.
Real, working exploits with reproducible PoCs. This is XBOW's home turf, and Keygraph matches it: No Exploit, No Report; a finding is filed only when a working proof-of-concept exists. Validation, not prediction, on both sides.
Keygraph drives remediation: source-level fixes proposed as reviewable pull requests, tracked to closure on one deduplicated, Jira-synced findings list with SLA tracking, rather than handing your team a report.
XBOW can re-test a fix when you request it. After the fix ships, Keygraph re-runs the exploit and re-scans to confirm the vulnerability is gone for good, then closes it in the system of record.
XBOW proved that autonomous exploitation works at scale, and topping HackerOne's US leaderboard against production targets is genuinely hard to do. Keygraph fields the same capability: autonomous testing validated by working exploits. On pure offense, the evidence is public.
In Doyensec's Aikido-sponsored study of Photoview 2.4.0, a $4,000 XBOW assessment reported 7 verified findings in about two days (Aikido's $4,000 scan reported 32). Keygraph ran Shannon 3.0, the open-source pentester the Keygraph platform is built on, against the same deployment: with Claude Opus 5 it found 23 verified issues in 2h 24m for $115 in model tokens, and a $6.10 run on DeepSeek v4 Flash still caught the critical pre-auth SQL injection (CVSS 9.8) that Photoview later patched. Read the full benchmark.
But as offensive AI matures, landing an exploit stops being the differentiator. What separates products now is everything around it: assessing the attack surface a single test never reaches, driving remediation to a reviewed, merged fix, and re-testing to confirm the issue stays closed. That's where a platform pulls away from a pentest-only product.
| Capability | Keygraph | XBOW |
|---|---|---|
| Autonomous offense | ||
| Autonomous pentest of the running appReal attacks at machine scale | ✓ | ✓ |
| Black-box testingExternal attacker view, zero knowledge | ✓zero-knowledge agents attack from outside, no code access | ✓its core discipline; external attacker view |
| Gray-box testingBlack-box plus provided context and steerability | ✓optional credentials, focus areas, business context, and OpenAPI specs steer the agents | ✓optional source upload and context steer its attacks |
| Exploit validation / reproducible PoCReproducible, not theoretical. Penetration test evidence. | ✓ | ✓ |
| Coverage & correlation | ||
| White-box (source-aware) testingAgents read your source | ✓agents read your source and generate precise exploits | Partialblack-box by default; optional source upload as context, not a source analyzer |
| Source-side SAST / SCA / secretsThe code-side surface | ✓the full source-side suite | ✕not marketed (offense only) |
| Static-dynamic correlationThe source map directs the pentest | ✓the static map directs the pentest | Partialmerges uploaded source with dynamic testing; no documented persistent correlated record |
| Business-logic testingInvariant discovery + PoC | ✓source-informed invariant discovery + PoC | Partialblack-box exploration may surface some |
| Deployment & trust | ||
| Runs air-gappedEngine runs in your perimeter | ✓on-prem or self-hosted in your own perimeter; Keygraph never receives your source or scan results; inference goes straight to the model endpoint you configure, so a full air-gap needs a local model endpoint | Partialhosted; self-hosted or air-gapped deployment not publicly documented |
| BYOK: model traffic stays yoursYour key, your gateway | ✓your own key, model, or AI gateway, including a self-hosted LiteLLM gateway, with usage tracking | ✕not publicly documented |
| Open core, remediation & pricing | ||
| Open sourceRead it before you buy | ✓Shannon, AGPL-3.0, 48k+ stars | ✕proprietary platform |
| Verified-patch remediation + system of recordFix, prove, track to closure | ✓fix → re-scan to prove → labeled PR; one canonical, deduplicated, Jira-synced findings list with SLA tracking | Partialoffense-centric: find → validate → document → suggest fixes → re-test on request |
| Pricing (public, sourced)What you actually pay | Pro from $50/developer/month (managed); Shannon (OSS) free; Enterprise custom, flat annual | usage-based pricing by quote; no public list price (the per-test Lightspeed offer was retired in July 2026) |
Swipe to compare →
XBOW prices by usage and quotes each environment, so cost scales with how much you test; the one public figure is the $4,000 Doyensec paid for an XBOW Plus assessment. Keygraph is one per-developer price for continuous testing across every app, with source-side scanning XBOW doesn't include.
Pricing sources: XBOW no longer publishes list prices; it prices by usage and quotes each environment (xbow.com/pricing, checked September 18, 2026), so your quote may differ. The XBOW figure uses the $4,000 per assessment Doyensec paid for XBOW Plus in its Aikido-sponsored study (see the benchmark). Keygraph Pro from $50/developer/month, a floor price. Estimates for comparison only.
Offense is one job
XBOW is a specialist: autonomous exploitation of the running app, at machine scale. Keygraph runs that and the rest of the program: continuous detection across the whole surface, findings tied to your codebase, and a loop driven to closure. A pentest finds exploits; an AppSec platform runs the program.
Exploitation, then closure
XBOW hands you validated exploits and a report, and re-tests on request; writing the fix and tracking it to closure are still your team's job. Keygraph drives remediation and re-exploits to verify: find → prove → fix → verify in one place, so the finding that goes in is the closure that comes out.
The whole surface
A vulnerable dependency, a leaked secret, a credential buried in an old commit: none of these surface as a running-app exploit, and an exploitation engine isn't built to scan for them. Keygraph covers the code-side surface as well as the layer an attacker reaches at runtime.
Close the loop, in your perimeter
XBOW is offense delivered as a hosted service. Keygraph runs inside your own environment, self-hosted or air-gapped, so Keygraph never receives your source or findings, and with a local model endpoint nothing leaves your perimeter at all.
Keygraph's white-box agents read your full source to generate precise exploits, alongside a black-box pentester with zero code access. XBOW is black-box by default and can ingest source to steer its attacks, but runs no standalone white-box analysis of your codebase.
Learn moreKeygraph also runs CPG and LLM SAST, reachability-ranked SCA, and secrets analysis. XBOW is offense-first and doesn't offer a source-side scanning suite (SAST/SCA/secrets).
Learn moreKeygraph's static map directs the pentest, so a proven-executable CVE or a static IDOR becomes a targeted exploit, kept as one correlated record across scans. XBOW can use uploaded source to guide an individual attack, but doesn't maintain a persistent source-to-runtime record.
Learn moreThe Keygraph platform is built on Shannon, our open-source AI pentester (AGPL-3.0, 48k+ stars), so you can read its code and run it on your own app before you buy. XBOW is a proprietary platform.
Learn moreClaims about XBOW on this page are drawn from the primary references below, verified September 18, 2026.
Run Keygraph against your own environment and judge the evidence: working exploits against your running app, reproducible proofs-of-concept, and fixes re-tested and proven closed.