Safety Controls & Daily Limits
Growth without gambling your account. Plan-based limits, human-paced timing, live logs, and review options where they fit each agent.
Key takeaways
- Every agent drives a real Chrome profile on your machine with your own login — no shared cloud IPs, no credentials on our servers.
- Volume is a range you set, clamped by your license plan, and a random target is drawn inside it each day so the footprint is never identical.
- Pacing is randomized per action (12–25 min at the safest, 5–18 steady, or clustered bursts) with forced 15–30 minute smart breaks on top.
- Review options are matched to risk: comment preview on LinkedIn, reply-stops-everything in Campaigns, Slack approve-before-post, reel approvals, Reddit age/karma gates.
- Full visibility: a plain-language live log, per-session counters, a persistent History tab, and a fleet status view per tool.
- Agents run in separate Chrome profiles so parallel automations can't collide, and each tracks its own budget.
Automation is only worth using if it doesn't put the account it's supposed to be growing at risk. That sounds obvious, and yet most tools in this category optimise for volume — because volume demos well and the consequences arrive later.
Every SocialKaptan agent shares the same foundation: your own browser on your own machine, irregular pacing, limits you can see and change, and a log of everything that happened. What differs per platform is the specifics, because LinkedIn, Instagram, and Reddit fail in different ways.
1. Local execution
This is the load-bearing decision and everything else builds on it. Each agent launches a persistent Chrome profile on your machine, you log into the platform yourself once, and the session stays there. Your credentials never touch a SocialKaptan server, because there's no server-side login to touch.
The alternative — a cloud tool signing into your account from its own infrastructure — puts your session on shared IP ranges that platforms already associate with datacenter traffic, and ties your account's fate to every other account on that provider. When it gets flagged, everyone gets flagged simultaneously.
A second benefit: agents get their own separate profiles. The LinkedIn Comment Agent, Campaigns, and the Post Generator don't share a browser window, so they can run in parallel without stealing clicks from each other.
2. Volume: ranges, not numbers
Every agent's limits are ranges you edit, and each day a random target is drawn inside them. An account doing exactly fourteen comments every single day is identifiable from the pattern alone, regardless of what those comments say.
| Agent | Default daily range |
|---|---|
| LinkedIn Comment Agent | 8–14 comments, 16–20 likes (likes kept 4–8 above comments) |
| LinkedIn Campaigns | 8–10 comments, 20 connection requests, 10 intro DMs (sequence-driven) |
| Instagram Comment Agent | 8–14 comments, 16–28 likes (gap of 8–12) |
| Instagram Campaigns | 5–10 DMs, plus 6–12 comments and 14–24 warm likes |
| Reddit Agent | 3–7 comments, 0–1 posts, 6–14 upvotes, 0–2 product mentions |
| Post Generator | 1–10 posts, spaced by an hour range you set |
| Reel Generator | 1–10 reels, spaced by an hour range you set |
The app labels these defaults as tested and recommends keeping them. You can raise them, up to the hard ceiling your license plan sets:
| Plan | Posts/day | Campaign prospects | ||
|---|---|---|---|---|
| Starter | 15 comments + 30 likes | 15 comments + 30 likes | 3 per platform | 5 |
| Pro | 30 comments + 60 likes | 30 comments + 60 likes | 6 per platform | 15 |
| Agency | 60 comments + 120 likes | 60 comments + 120 likes | 10 per platform | 40 |
Notice that likes always sit well above comments. Likes are the cheap signal; comments are the expensive one. An account that comments on everything it touches looks nothing like a person scrolling a feed, so the gap is enforced rather than left to chance.
3. Pacing: no rhythm to detect
The most common giveaway isn't volume — it's regularity. On LinkedIn and Instagram, the Speed setting controls how actions are spaced, and the default varies it automatically based on time of day.
| Profile | Gap between actions |
|---|---|
| Recommended (natural pace) | Chosen per action, weighted by time of day |
| Slowest — safest | 12–25 minutes |
| Steady — even spacing | 5–18 minutes |
| Fastest — bursts | 2–5 minutes, then 15–25 minutes |
On top of whichever profile is active, a smart break every few actions forces a 15–30 minute pause that can't be disabled. The first action of a run also waits a short random delay, so nothing starts on the exact second you press the button.
Reddit goes further, because Reddit is stricter. Each day the agent rolls a behaviour archetype — rest, light, normal, heavy, or binge — which sets that day's volume; activity is confined to a randomized waking-hours window; comments happen in micro-sessions of one or two; pauses are drawn from weighted tiers ranging from seconds to two hours; and any idle stretch over five minutes closes the browser entirely and reopens it later.
4. Review, matched to risk
Different actions deserve different amounts of human oversight. Publishing to your own feed is low-stakes; DMing a stranger is not. So the review options aren't uniform:
| Agent | Review option | Default |
|---|---|---|
| LinkedIn Comment Agent | Preview only — types the comment without submitting, on single runs | Off (opt in) |
| Instagram Campaigns | Stops for a person the moment they reply, and never puts a link in a DM | Always on, not optional |
| Post Generator | Slack approve-before-post — drafts wait in your channel for Approve / Reject | Off (needs Slack) |
| Reel Generator | Require approval — Slack, or an Awaiting approval queue on your machine | Off (opt in) |
| Reddit Agent | Age and karma gates block posting and product mentions per subreddit | Always on |
| LinkedIn Campaigns | Stops the moment a prospect replies — remaining steps are skipped and they appear in Replies | Always on |
Campaigns are the clearest expression of the philosophy: there is no setting that lets one keep messaging someone who has answered. The moment a prospect replies, that person is dropped out of automation and handed back to you — the conversation a human should be having stays a human decision.
5. Visibility
You can't judge whether automation is behaving if you can't see it. Four layers cover this:
- The live activity log narrates each step in plain language — which post it opened, what it liked, how long it's waiting next — with a developer view for raw output when you want it.
- Session counters tally the actions taken in the current run: comments, likes, DMs, posts.
- The History tab keeps a persistent record per agent that survives restarts, so "what did it do overnight" always has an answer.
- Pipeline views — the People tab on each campaign and on the DM Agent — show the exact status of every person, and guarantee nobody is approached twice.
Runs also report their outcome back to a fleet status view, so a tool that's failing across many machines (usually because a platform changed its UI) shows up as degraded rather than as a mystery on your desk.
6. Things the agents refuse to do
Some safety comes from restrictions rather than settings:
- No campaign step ever messages someone who has already replied.
- No links in a first Instagram DM, and no pitch in a LinkedIn connection note.
- No more than one follow-up if someone doesn't reply — then they're retired, not re-queued.
- No re-approaching a profile already visited, skipped, or messaged.
- No Reddit posting or product mentions in a subreddit whose age or karma gate you haven't actually cleared, and no free-karma or vote-exchange communities.
- No stale LinkedIn invites lingering — pending requests expire after three days and the person moves on.
What this doesn't promise
No tool can guarantee zero risk. Platforms change their rules and their detection without notice, and automation of any kind is against the terms of service on most of them. Reddit in particular enforces community rules aggressively and unevenly, and no amount of pacing substitutes for reading a subreddit's rules yourself.
What these controls do is remove the failure modes that are actually within a tool's power to avoid: shared IPs, mechanical timing, aggressive volume, invisible behaviour, and messaging people who never should have been contacted. The rest depends on how you configure it and what you use it to say.
A conservative starting configuration
- 1Leave every default limit exactly where it is for the first two weeks.
- 2Keep Speed on Recommended, and keep the browser visible so you can see what's happening.
- 3Run each agent once, manually, and read its output before enabling any all-day mode.
- 4Turn on the review checkpoints available to you: comment preview, Slack approval, reel approval.
- 5On a new Reddit account, run upvote-only for several days before commenting.
- 6Check the History tab after the first full day, not the first hour.
- 7Raise limits only after a couple of clean weeks — and one agent at a time.
Frequently asked questions
Each agent has editable daily ranges, and a random target is drawn inside them each day. Defaults are 8–14 comments and 16–20 likes on LinkedIn, 8–14 and 16–28 on Instagram, 5–10 DMs with 6–12 comments and 14–24 likes in Instagram Campaigns, 8–10 campaign comments with 20 connection requests and 10 intro DMs on LinkedIn, and 3–7 Reddit comments with 0–1 posts. Your plan sets the ceiling above those: 15/30 per platform on Starter, 30/60 on Pro, 60/120 on Agency.
Because a fixed number is a signature. An account that does exactly the same volume every day is identifiable from the pattern alone, no matter how good the content is. Drawing a random target inside a range — and, on Reddit, rolling an entire day archetype including genuine days off — makes the footprint look like a person's.
On a LinkedIn Comment Agent single run, preview types the AI comment into the box without submitting so you can read it and decide. Campaigns take the opposite approach on both platforms: they send inside the daily limits you set, and stop for a person the moment that person replies.
Yes, four ways: a live plain-language activity log while it runs, session counters for the current run, a persistent History tab per agent that survives restarts, and pipeline views (People, Replies) showing the exact status of every person Campaigns and the DM Agent have touched.
Local Chrome profiles, randomized pacing, editable limits, live logs, and History apply everywhere. What differs is volume and review: Instagram DMs are capped in single digits and never carry links, Reddit adds age/karma gates and day archetypes, the Post and Reel Generators offer Slack approval, and Campaigns stop for anyone who replies.
Likes are low-risk and comments are the expensive action, so a minimum gap between them is enforced — 4–8 extra likes on LinkedIn, 8–12 on Instagram, randomized daily. An account that comments on every post it touches doesn't resemble a real person browsing.
Each agent uses its own Chrome profile and tracks its own budget, so they don't collide. Run All starts connected agents' daily modes together; Campaigns are started from the Campaigns screen. Worth knowing: running the Comment Agent and Campaigns together means two separate comment budgets are in play, so account for that before raising either one.
Nothing already posted is affected, and progress is checkpointed — restarting won't double-post or double-count against the day's target. The History tab keeps the record either way.
No — the hidden option moves Chrome off-screen rather than using a detectable headless mode. But it does make problems harder to notice, so it's worth leaving the browser visible until you trust a agent's output.
No, and any tool that claims otherwise is worth distrusting. Automation is against the terms of service on most platforms and detection changes without notice. What's controllable — where the session runs, how irregular the pacing is, how conservative the volume is, whether you can see and stop what's happening, and whether the same person can be approached twice — is what these controls address.