Skip to content
KitploitKITPLOIT
ToolsExploitsBlog
Log in
Submit
ToolsExploitsBlog
Submit

Hacking, PenTest, and Cybersecurity Tools for Your Security Arsenal!

Kitploit is a directory of hacking, cybersecurity, and pentesting tools. Discover the latest project updates to find vulnerabilities, analyze systems, automate testing, and strengthen your security.

FeedsContactPrivacy© 2026 Kitploit

Tool Directory

Categories

View all categories
Loading categories
ReAgentLive — Let AI agents explore a running game from the inside. Live, safe, instant. | Kitploit
Tools/GitHubGitHub/krixx1337/reagentlive
Dynamic Analysis (Sandboxing)Memory ForensicsReverse EngineeringScripting & AutomationDebuggersUtilities & FrameworksBinary AnalysisAI-Assisted ReversingAI Security
GitHubkrixx1337/reagentlive

ReAgentLive

Let AI agents explore a running game from the inside. Live, safe, instant.

613h 48m agoNot yet reviewed

Most Popular

View all →

Discover the most used tools by our community.

Explore all tools

Browse our collection of tools

View all tools →
View Repository
Share

ReAgentLive

The live, in-process probe for AI-driven game reversing.

Static tools tell you what the code does. ReAgentLive tells you what is true in the running game, right now, and lets you test it without breaking anything.

Live. It runs inside the game and answers from live memory, on the game's own tick. Safe. References that know when an object died, writes that check first, changes that undo themselves. Immediate. No database, no hours of analysis: questions about code are answered from live bytes in milliseconds.

First-class Unreal Engine support. Build the Unreal pack against the SDK Dumper-7 generates for your game, and work by name: find objects, read and follow reflected fields, call functions, list every instance of a class, all on the game thread. Everything else (memory, code, native structs) works in any game.


The problem

Point an AI agent at a running game with the usual tools and it hits the same walls:

  • Reads from outside are torn. Two values read a moment apart never came from the same frame.
  • Pointers go stale. An enemy respawns and yesterday's address is now someone else's memory.
  • Raw writes crash things. One wrong byte, no undo.
  • Static analysis is slow. Hours of whole-program analysis before the first answer, and again after every patch.
  • Huge tool surfaces eat context. A hundred tools cost an agent its budget before it does anything.

ReAgentLive is built around those walls.

30 seconds of it

Load one DLL into any x64 process, here a running Python interpreter, and ask. No game, no SDK, no database. Output condensed from the JSON the console prints:

> reagentlive -c "search text PYTHONPATH --module python313.dll --paged --timing"
  hits: python313.dll+0x2c1ca8                               4.9 ms

> reagentlive -c "xrefs python313.dll+0x2c1ca8 --timing"
  lea  python313.dll+0x12ff02  in function +0x12fd64..+0x1301ce
  scanned all of python313.dll                               29.6 ms

> reagentlive -c "disasm python313.dll+0x12ff02 3"
  +0x12ff02  lea rdx, [0x7FF827981CA8]   string: "PYTHONPATH"
  +0x12ff09  mov rcx, rbx
  +0x12ff0c  call python313.dll+0x1301d0

A string, every place the code uses it, and the code around it: three commands, well under a second, straight from the bytes the process is running. Every address comes back as module + rva, ready to paste into your disassembler.

Why it is different

No single one of these is new. The combination is.

ReAgentLiveThe usual way
Inside the process, on a thread the host trusts. No debugger attached, no hooks of its ownA debugger, a hooking framework, or reading from outside
Tick-accurate: on the engine's thread, everything in one answer comes from the same momentTorn reads from outside, or polling
Safe by construction: references that refuse a replaced object, writes only over the bytes you expected, pins and trials that expire on their ownRaw writes handed to an agent
Shaped for an AI's costs: one console, few commands that each do a whole lookup, filtered output, state kept between runsLarge tool surfaces that fill the context first
Immediate: one-address code questions from live bytes, no databaseHours of analysis before the first answer
Reflection and raw memory in one tool: Unreal objects and native structs side by sideOne or the other

Low footprint

ReAgentLive is quiet. It answers when asked and otherwise stays out of the game's way:

  • No hooks of its own. The core never patches code or swaps a pointer; where the tick comes from is the host's business.
  • No threads of its own. It runs on a thread the host already has. The own-thread example adds exactly one, its tick; background search adds a worker only when the host allows it.
  • No process handle. The client never opens the game process: no OpenProcess, no ReadProcessMemory, no debugger. Requests travel through one named shared-memory section.
  • Next to nothing while idle: one atomic increment and one compare per tick. Fixed buffers, no allocation at load.

Low footprint means it does not disturb the game. It does not mean hidden: see what it is not.

What you can do

If you want toReAgentLive gives you
See what the game is doing right nowfind objects, dump them, follow field paths, read containers, sample any value every tick
Test an idea without rebuildingguarded writes, pins that hold a value, trial writes that undo themselves, engine calls
Understand code without waitingthe fast lane: instructions, references, functions using an offset, signature checks
Name what you foundlabel unreflected members once; export labels to your static tool
Let an AI do all of the aboveone console with JSON in and out, a Python client, and a skill that teaches the method

Try it in five minutes

The own-thread example is a DLL that works in any x64 process: no mod, no SDK, no game knowledge.

# From a Visual Studio x64 developer shell, in the ReAgentLive folder:
cmake -S examples/own_thread -B build/own-thread -G Ninja -DCMAKE_BUILD_TYPE=Release
cmake --build build/own-thread
# Load reagentlive-own-thread.dll into the target with any DLL loader, then:
reagentlive -c "status; modules"

You get memory, the fast lane and native objects (vtable, pointer chains, region, strings). With no pid the console finds the process on its own. More: own-thread example.

Best on the game thread

The own-thread DLL is the fast way in. ReAgentLive is at its best when the host ticks it on the engine's game thread, from a mod or a hook you already own:

Host ticks onYou get
The game threadeverything: object and field commands, engine calls, and a tick that is one moment of the game (batch and get say with same_tick whether they fit in one)
Another thread (its own, or the frame present)memory, the fast lane and native objects; engine commands are refused, and reads in one tick need not line up

status says which one you have (host_thread). How to get onto the game thread is the host's business; what each choice gives is in hosting.

Go further

  • Unreal games: the Unreal pack over your Dumper-7 SDK adds objects, reflected fields, function calls, census (what exists, what just spawned) and each (one field across every instance, in one tick). Unreal pack
  • Your own mod as the host: tick the agent from the game thread and get all of the above. Hosting
  • Logic outside the game: keep a tool or a mod's logic in its own program and use the core as a lean in-process proxy. Outside programs
  • AI agents: the reagentlive skill teaches the investigation method: rules, workflow, pitfalls. Skill

Where it can grow

The core knows no engine. Everything engine-specific lives in a pack: today one, for Unreal. Memory, the fast lane and native objects already work in any engine; a pack adds objects, fields and calls on top.

Natural next packs are Unity (Mono and IL2CPP) and Source 2. They are candidates, not promises: one gets built when a real project needs it (backlog).

Plays well with others

Download Tool