Autopilot, copilot, manual
Deploy agents that inherit model/settings state, reason through scoped tasks, call tools, and expose current/last tool status in the UI.
exploitbot is a native macOS beta for authorized security testing: local MLX models, full-agent tool use, CVE intelligence, supply-chain workflows, live logs, and report-ready evidence.
The beta focuses on full-loop usability: broad tool discovery, precise CVE/context retrieval, visible tool progress, persistent settings, and local model runtime proofing.
Deploy agents that inherit model/settings state, reason through scoped tasks, call tools, and expose current/last tool status in the UI.
Recon, web, network, credentials, exploit, post-exploit, OSINT, supply-chain, CVE, context, and shell execution are available through one schema surface.
Search local CVE data, import CVE lists, include only the rows you care about, and inject focused context instead of dumping whole histories into prompts.
TruffleHog, Syft, Grype, OSV Scanner, Nuclei, CVE search, and shell fallback are wired into the same action/status lifecycle.
Qwen hybrid SSM attention and MiniMax full-KV attention paths are validated with TurboQuant q4 KV, prefix cache, paged cache, block L2, and warm replay checks.
Keep artifacts, findings, terminal paths, screenshots, CVE rows, and report state tied to the engagement instead of scattered across ad hoc notes.
Screenshots are from the latest app proof passes and cover agent status, tool tabs, CVE import, runtime settings, stash, reports, and terminal/tool routing.

Queued, running, done, and error states are visible in chat, sidebars, and agent panels.

Subdomain, DNS, HTTP probing, and service discovery routes feed the operation context.

Search, prepare, and execute flows retain status and evidence instead of hiding tool progress.

Paste or import lists, include specific CVEs, and keep CVE status visible in settings and tools.

Parser, reasoning, cache budgets, KV quantization, prefix cache, and model settings persist across runs.

Tool availability, installers, terminal paths, and execution status are kept in one operational surface.

SMB, SNMP, capture, tunnels, and protocol workflows can be queued from their native tab surfaces.

Hash identification, cracking, online checks, and secret scans stay attached to the engagement timeline.

Metadata, screenshots, username checks, and email checks produce previewable evidence.

Findings, chains, impact, and artifacts flow toward Markdown, HTML, PDF, and JSON reports.
The beta DMG bundles the Python engine and a vMLX-compatible Python runtime, then verifies signing, notarization, and core runtime capabilities.
MXFP4-MTP lanes validate hybrid cache components, SSM companion state, prefix cache, and repeat-prompt hits.
MiniMax proofing covers full-KV replay, TurboQuant q4, parser selection, thinking/no-thinking paths, and cached repeat completions.
MiniMax JANG_K is metadata-verified for parser and cache wiring; live 80 GB load testing is reserved for a quiet machine.
macOS 14+ on Apple Silicon. The app runs local inference and can install/use external security tools as needed.