Agentic operations
An AI operator with the same tool surface you have, with full audit and instantly revocable scope.
Connect any MCP-compatible client (Claude Code, Cursor, Codex) to the ApeX teamserver. The teamserver exposes 25+ structured tools across 15 scopes. Scoped tokens let a client-reviewed AI run read-only while you keep full autonomy for yourself. Every tool call is logged with arguments, agent id, result, and duration. The log is immutable. Sessions revoke in a single click.
The tool surface
| Category | Representative capabilities |
|---|---|
| Query | list/get agents, tasks, listeners, modules, payloads, audit log |
| Execute | run commands, Lua modules, .NET assemblies, aggressor commands on any agent |
| Manage | upload modules, create payloads, create/start/stop listeners, manage vault files |
| Documentation | embedded reference docs (Lua API, command reference, BOF conventions, profile format, aggressor SDK) |
A full engagement, one instruction
operator> start on APP-OWA1, map lateral paths to a DC without
mimikatz-style apis. stage findings, write a brief.
agent> reading Lua api references
done.
agent> drafting module recon_v1.lua
scope: hostname, domain, token privileges, open shares,
named-pipe surfaces, routing table
mode: non-destructive
agent> upload_lua_module(APP-OWA1, recon_v1)
ok, module id 2b19...9f03
agent> execute_lua_module(APP-OWA1, recon_v1)
domain: corp.local
dc candidates: DC01 (10.1.0.4), DC02 (10.1.0.5)
current context: svc_backup (service account)
SeImpersonatePrivilege: enabled
agent> drafting module lateral_plan_v1.lua
scope: enumerate reachable services on DC01, probe
for named-pipe impersonation via spoolsv
mode: non-destructive, timeout 20s
agent> upload_lua_module(APP-OWA1, lateral_plan_v1)
ok, module id 7f42...a1c6
agent> execute_lua_module(APP-OWA1, lateral_plan_v1)
DC01 \pipe\spoolss reachable from APP-OWA1
path identified: service-account impersonation via
spoolsv on DC01
agent> staging apex-vault/engagement-7xq2/brief.md
brief written. 3 findings, 1 recommended path.
dwelling for review in agent console.
The same console, with or without an AI
ApeX is operator first. The agents list, task queue, module catalog, and command palette are the same whether a human types the next step or an MCP-connected AI stages one for review. The console is the shared surface. The audit log does not care which hand pressed enter.
Modules the agent can reach for
Aggressor packages, Lua modules, and .NET assemblies are versioned and lifecycle managed in the teamserver. The AI operator lists, uploads, and runs from the same catalog you use. Havoc API compatibility means existing aggressor packages port with minimal changes.
Scoped tokens
Give a client-reviewed AI read-only access. Give yourself full autonomy. Tokens are per-user, scoped to exact tool categories, and revocable in one click.
Immutable audit
Every tool call is logged with arguments, agent id, result, and duration. The log is append-only. Not editable, not forgeable, exportable for client reports.
Human in the loop
AI stages the plan. You approve destructive actions in the agent console. Approval posture is set per engagement; strict review for client work, open loop for lab testing.
Grounded answers, not hallucinations
The LLM reads the Lua API, command reference, BOF conventions, profile format, and aggressor SDK from inside MCP. When it drafts a module, it quotes real function signatures. When it asks about a profile field, it reads the schema. Documentation is a first-class tool category, not a prompt-stuffing hack.