I've been trying to make my first matrix character and I've had a good look at the rules in SR4A and Unwired, looked over countless forum posts from the last ~7 years across various forums, and I'm still not sure I've got the rules for agents correct. Assume that all programs are cracked.

The rules indicate that only 1 agent from a commlink can be copied onto a node at a time, but that agent may be loaded with its own agents. Is there anything stopping me from just loading up 5 dummy commlinks with 5 agents each, loading 4 agents into 1, and copying all 25 onto a node?

When agents are loaded into a persona, or another agent, they're treated as programs and don't have their own access ID. Does that mean I can copy an agent from 1 commlink to another, and then load and run both agents on both commlinks' personas on the same node at the same time?

If I load up my persona or an agent with more agents, does each agent need its own copy of a program to use it or can multiple programs share the same attack/armor program?

Similarly, if I'm on a node which I'm accessing from multiple personas using multiple commlinks, or a node which has some of my agents copied onto it, can I designate and run programs for personas/agents running on different devices and vice versa? I'd assume if this was allowed that the programs would be vulnerable to spoof commands, and as far as I can tell it's not covered by RAW.

Also I can copy an agent onto a node, the agent functions with its own access ID, and henceforth functions independently, running from the node. Does this mean I can load up my persona with a bunch of agents (which are now running on my ID, which may be spoofed) and then copy an agent from that commlink onto a node (which may be loaded with more agents) and then run all of those agents in one node?

I don't want to create a full "agent smith" scenario, I just want to find a practical way to use ~5 agents at the same time. Every nuyen saved means I can diversify my character so they can fill a broader roll.

Yes I am aware that there are a lot of advantages to hacking things yourself and there are a lot of disadvantages to using agents for hacking.

The idea would be to have an AR/VR commlink loaded with all of my programs and a handful of agents (~3), with the agents running and deactivating programs for me, as well as using programs which don't have an opposed test, or otherwise do not suffer from the agents lowered stats, and simultaneously have another commlink loaded with agents and copies of certain programs (attack, browse, nuke, and scan for instance)) that I would use for mass probing stuff as well as flooding a node (via creating a reusable exploit) with combat agents to fight IC when matrix security detected me.

1) Agents count as programs run, so you're bogging down your own system while attempting to streamline performance.

2) Programs run by Agents are limited to the Rating of the Agent (SR4A, page 234, Agents and IC, Payload, 1st paragraph, 3rd sentence). The only way around this is the Optimization program option from Unwired.

3) Sounds like what you're trying to do is a botnet (pricing is page 93 of Unwired, rules are on page 100).

4) Also, sounds like you'd be better served using Compiling and Registering as a TM.


I'm having a hard time understanding what you're asking.

1, I know about the response rules, and they're a slap on the wrist. ESPECIALLY if its for a swarm of agents that are just there as essentially, a distraction. That being said the solution is simply to have less agents per commlink, and have more commlinks. As for bogging down the main commlink by running agents, I'd be running ergonomic programs for the "always on" programs (like biofeedback filter, reality filter, and stealth), and have the agents deactivate the ones I didn't need via scripts, essentially granting me extra matrix actions at the sacrifice of some response, or at least at the sacrifice of some response until the agents have deactivated the unnecessary programs (including themselves).

2, Yup, and I'm okay with that. Hacking is a secondary role of the character, and I'd be starting with a response 4 commlink, so agents would run programs just as well as I would. I'd start replacing my programs with optimized programs later in play.

3, A miniature, self contained botnet. With the added advantage that all of the devices running the net would be my commlinks. The idea being to be somewhat more subtle and focused than a botnet.

4, Hacking is a secondary role, so I'm not going TM.

And lastly, as on page 234 of SR4A under payload it says "If your agent is running in your node, you may run a program and
designate it for your agent’s use and vice versa."

My question is, does that agent need to be one running from your commlink node, or can you just designate programs for use by any friendly agent and vice versa as long as you're subscribed into the same node?

Anyway thanks for the response, I appreciate it!

Unless I miss my guess, only one copy of an agent can be on a node, period. So, an agent can't mass-replicate itself, like Agent Smith. You also can't copy an agent onto itself for the same reason, since they're still running on the same node.

What you can do is get multiple agent programs, and run them all at once, but it's very expensive and not especially effective.

Cain that matches my understanding. However the cost effectiveness of mass agents goes up if you're using Unwired, because you can buy hacked agents for 1/10th the price, and if need be you can copy the agent and rewrite their access ID, which is done with an extended test. That or just buy lots of agents for 1/10th the price. So for the price of buying 1 agent I could buy 10 hacked agents.

They actually errata'd part of that. You can't spoof an agents Access ID anymore, so that trick went away. The line developer at the time, Peter Taylor aka Synner, said it was a mistake to include that bit in the first place, it got through by accident. You can buy 10 hacked agents, but their rating drops very quickly, so it's not as effective as you might think.

The mass agent trick was first made popular here, on Dumpshock. A bunch of fans raised a stink over it, dubbing it the Agent Smith army, and made a few demonstrations about how self-replicating agents could munch Zurich Orbital for breakfast. There were a couple of tricks put into place as a result, including clarifying that agents started on your commlink, and the ones mentioned. The changes to the Teamwork test were also a result of Dumpshock finding mistakes in the system and pushing for fixes.

A slap hitting the wrist it was intended for . You are cracking the agent multiple times (or buying multiple cracked ones if you prefer), then setting each copy up with the programs it needs. System has to stay at the level of your agent programs, as optimization is not available for those. Consider Response 5.

So System (Analyze, Agent[Programs], Agent[Programs]). Analyze is there to defend against hacking. Optimization of the programs counts for both the Agent and the System running it - lowest limit applies.

Payload - no Filters, not Biofeedback, not Reality. I´d not even run Reality Filters on my own Persona. Agents don´t have a nervous system to detect, and Reality Filters are not worth the program slot. Analyze, Armor, Attack Program, Stealth.




If hacking is just a secondary role, I advise against starting with self-created micromanaged botnets. Might as well go all the way at that point, 8ish dice pools are not worth the effort.


The Agent and the Program have to be running on the same node. A subscription is one step below.



The trouble with this approach is that it leaves tons of traces, and you´ll never know what the good guys have figured out. It´s somehow the opposite of emission control. Agents are also stupid. Once you create scripts they can be triggered at the most inopportune moments.
Defensive Agents are obviously safer than offensive ones. What you can do is run IC for your PAN on the strongest devices. That strategy scaling up with available ressources is somewhat the bane of the SR4 matrix rules - everyone important could be doing it, and stuff would be hard for the hacker. Come to an agreement with your GM. Agents can be fun, but are a "sometimes food".

For me, if you start using a swarm of agents, you move to botnet rules, no matter if you're using an actual botnet or something you made with your own hardware and agents.

While Synner and I never saw eye-to-eye on much, he did have a very strong dedication to the game. He realized early on that self-replicating agents would doom the matrix, so he listened to us fans, and slammed down on it hard.

Infinite agents cause a huge number of problems. First of all, that many agents in cybercombat slows everything to a crawl. I never encountered a real Agent Smith army, but I did have someone have agents replicate themselves until he had twenty or so backing him up. The only way to cope with that is to have twenty agents on defense, and rolling forty attacks per round ruined the game. It was unmanageable. A real Agent Smith army could have hundreds of thousands of agents, battling hundreds of thousands of self-replicating IC. So, one cybercombat could spill over and crash the entire Matrix.

But, let's look at why one replicating agent is so much worse than ten pirated ones. One agent could sneak onto a Node or Nexus, and then copy itself so many times, the entire server would come screeching to a halt. It didn't need any other programs, or to really take any actions that would get it noticed. Ten different agents, however, might not even slow down a Nexus, especially if the persona limit is big enough. They'd have to start hacking things to affect the system, which would probably get them noticed; and since the Teamwork test was fixed, they wouldn't be scarier than a normal decker.

How was the Teamwork text fixed? I don't recall off the top of my head

Thanks for the clarification guys.

The goal I'm going for isn't "crash things with buttones of agents", it's "hack, and use a handful of agents to supplement my hacking".

There's going to be a clear gap between my main hacking commlink, and a commlink full of assault agents, which I fully expect to be traced when I use it.

The overall strategy was to use the main commlink for hacking with exploit, spoof, and stealth being the primary hacking tools. As soon as I had access I'd create a reusable exploit while I had agents do stuff like analyze icons, do matrix perception tests, and run the programs I needed and deactivate the ones I didn't, and then I'd just hack as normal. As soon as an alert was triggered I'd abandon all pretense of subtlety and hack the node with the bot commlink, see if I could finish the hack, if I could finish it while tasking the agents to make a mess to cover me, if not, just jack out and turn off wireless while the bot commlink fights IC and what not.

I've always found that it is better to just drop a few high-end Bait Worms into the system when alarms go off and let them draw all the attention while you try to continue on mission quietly and hidden. Though I am a fan of decent IC on the Comlink for Defense.

That's pretty much what the commlink full of agents is for, but with more versatility.

I've noticed that if there isn't a response within a couple days, if at least one person posts a response then it opens up a floodgate of other posts. So in part my first post was to try and answer your questions, but it also was in order to cause more people to post.

In any event: have you looked at my Vehicles, Drones, and Agents thread? It covers a few things, including the dogbrain agent scripts, and shows how you can actually get a decent hacker for just money.

What is your program load on the commlink full of agents?

Depending on the rating it was going to be 3-5 agents with analyze and attack, and possibly armor. I eventually moved to a character concept with only 1 hacking commlink... for now, but I'll probably get one in game.

So I'd be looking at 9-10 programs counting agents all up.

The problem with this setup is the edge acutally goes to the defender; its easier to lock down a node than it is to hack it.

Fun thing about alerts: They carry a firewall bonus; and also strip the user of all rights(per unwired), so getting spotted going for administration is a bust, and you can't just turn the alarm off afterwards. Additionally, the node will resist changes, and a potential firewall of 10 is hard to beat with dogbrains.

On top of that, its easier for alert defenders to lock out chokepoint nodes; particularly when it comes to data bombing arriving logins.



just as an FYI, each program in the payload counts as a running program, so 9-10 programs per agent adds the heck up. Payloads aren't a free pass to run whatever.

Teamwork tests were originally unlimited. So, you could get however many dice from your team as they rolled, there was no cap. This didn't look like a problem in practice, since there's only so many PC's in a team, with maybe a few NPC's.

Unlimited agents was the killer. If you replicated an agent 100 times, you could have 100 assistants rolling to help you. Since in the 4e core, the toughest system could be hacked with 18 successes, this was extreme overkill. There was no way to stop it, which is why Teamwork tests needed the fix.

With full ergonomic use a R4 node can run 8 programs before being degraded. Three agents with two ergonomic programs each count as 9 programs, ergonomics can discount that to 5. Two agents running three count 8(4), a better point to be at.


The first agent you should run is inside your persona for tasks you don´t want to monitor constantly. Scanning for attackers while you do meat-world stuff, running Browse, whatever. The second is usually some form of dedicated IC running on a different device inside your PAN. My group has agreed to stop the arms race regarding constantly running IC at that point. (A forced push to the botnet rules like Blade suggests is good, we have it as "you are not allowed to make the GM smile").

While true, agents without a swiss army knife configuration are pretty vulnerable and/or dumb. For example, no analyze means they fall prey to any defender with stealth.

Also optimization has a limit to how far it goes, and (away from books now) iirc its based indirectly on system, or whichever stat governs processor limit. It effectively gives you one more processor limit of programs before degredation. Less useful if you want to run tons of agents, very useful on a nexi.

Yeah that's pretty logical.
Shame too, I was looking forward to going with a themed set of 3 agents to micromanage my programs.

Although technically if I run 1-2 from my persona and 1-2 from another commlink it's doable.

Three agents at R4 is far away from using the "MORE IS BETTER!!1!" strategy, and if you really deliver the theme it will be ok. Dice pools of 10 do not ruin the game at low numbers (R4 agent with R6 ergonomic optimized Analyze, Stealth, true payload program). Any technomancer is likely to have more and better sprites. What themed agents would you like to run?

IE: Start with one R6 agent for your persona having access to all programs you don´t currently need yourself, run a R5 secondary node/commlink for two other (max effective) R5 agents. The primary one is versatile, the secondary ones have a payload of up to 8 programs total before their rating drops to 4, and consequentially pools of 11-10. If you use that for matrix combat, your GM will increase the number of IC, if you use it to control nodes everything might be fine. Using the matrix rules well is much about establishing a balance of power for your group, as influencing the numbers is pretty cheap.

The agents would be themed as parts of a classic D&D party.

The warrior was going to run programs for matrix combat, the wizard was going to run programs for hacking and accessing nodes, and the thief was going to run programs for hacking and manipulating data.

So the thief and the wizard would help me hack by running secondary processes such as running matrix perception tests while I'm exploiting things, or corrupting data while I'm decrypting something, and I'd run the warrior when IC got involved.

(I just corrected my use of the "optimization" option vs the "ergonomic" one in a post above, my apologies.)


As you want to specialize your agents and fight with only one of them a better commlink might be sufficient, as a Response 6 chip would allow a System of 5 with 15 programs, enough for wizard/thief running inside your persona, using your programs, and the warrior taking up about four slots, running on your node but not in your persona. (different icon, different damage monitor).
Note that you save a few slots because wizard/thief don´t run stealth on their own or Analyze on principle, but any matrix damage they take goes to your personas icon, and any system log leads back to you. They are basically just "AI modules" for your programs.

Yeah the idea was to run those four as "AI modules" to run my programs for me, and before I scrapped it, the secondary commlink would have had the independent agents.

You need to look at Fetch modules, in unwired.

Its basically a USB dongle that contains an agent. BUT, it basically exists to give that agent extra memory for programs while using the response of the machine its plugged into.

Seen the fetch module, but the RAW for the fetch module is that it's only capable of running browse, and I don't want to torment my GM by making him homerule stuff.

Maybe Clustering or Slaving is the answer?

I think I'm going to do it with one commlink for now and take the response hit, upgrade it in play, buy a secondary commlink and slave that to the first one. After all my character has a sleep regulator, skills to code with, and matrix contacts, so I'll have time to get better programs, either coding them myself, or bartering for them off the 'net.