Mirgent ships four specialist AI agents — Content Planner, Lead Hunter, Analytics, and Engagement — that run on schedules you set, against rules you define. They draft, scan, report, and engage; every meaningful action lands in your approval queue. You direct, AI executes, you approve.
Mirgent agents are designed to make you faster, not replace you. Six features that keep humans in the loop without adding friction.
Each agent runs on a cron expression you set — daily 9am, every 4h, weekdays only, whatever fits your content cadence. Pause any agent with one click without losing its memory.
Agents accumulate longitudinal memory: which posts performed, which leads converted, which keywords burned. They reference the memory on every run so quality compounds.
Agents are nodes in your workflow builder. Chain them: Content Planner → human approval → cross-post fanout → Analytics weekly digest. No code.
Hard rule: every agent action that creates public content (post, comment, follow) routes through your approval queue. Per-agent override only with explicit consent.
Agents read your business context (product, tone, positioning, banned phrases) before every output. The same agent draft for two different products produces two different drafts.
Every agent run is logged with input, output, cost, and outcome. Replay decisions, debug regressions, prove compliance.
Each agent owns one job and does it well. Run any subset — Free plan includes one agent slot, Pro unlocks all four plus custom-built agents.
Drafts platform-tuned posts on your schedule. Reads your RSS feeds, business context, and recent performance; outputs drafts you approve. Pairs with Cross-Post for one-click multi-platform publishing.
Scans 9 platforms (Twitter, Reddit, LinkedIn, Quora, Hacker News, YouTube, TikTok, Mastodon, Bluesky) hourly for buying-intent threads. Ranks by AI confidence, drafts replies for your approval. Cooldowns + per-platform quotas built in.
Runs weekly. Pulls performance metrics from each connected platform, identifies what worked, and emails a digest to your inbox. No staring at dashboards.
Replies, likes, and follows within rules you set: max 5 actions per platform per day, only on threads matching your filters, never twice to the same user. Budget-bounded so it can't go rogue.
Agents are not magic. Every run follows the same pipeline; understanding it makes the system predictable and debuggable.
At each scheduled time, Mirgent evaluates the agent's active goals (e.g. 'post 3 LinkedIn drafts daily', 'find 10 SaaS leads on Reddit'). Goals carry budgets — token caps, action caps, time windows — that the agent cannot exceed.
The agent loads its longitudinal memory (last 30 days of inputs, outputs, outcomes) plus your business context (brand, tone, positioning). This is the prompt-cache sweet spot: identical context across runs means cheap LLM calls.
The agent calls a small set of tools (post draft, lead scrape, analytics fetch, engagement check) in a loop. Each tool call is bounded: max retries, timeout, error fallback. The agent stops when goal is met or budget is exhausted.
All public-content outputs go to your queue (drafts, lead replies). Internal outputs (analytics digest, engagement counts) ship directly. Every run adds an episode to memory: action + outcome, so next run learns.
Mirgent does not resell tokens. You bring your OpenAI, Anthropic, or other provider keys; we orchestrate the calls.
Three concrete patterns that map to real businesses we've helped run.
Content Planner drafts daily LinkedIn + Twitter posts from your blog RSS. Lead Hunter surfaces 5 high-intent threads per day. Analytics Agent emails a Friday recap. Total time: 15 minutes per day approving the queue. Without Mirgent: 2 hours.
One agent profile per client. Each profile carries that client's business context, brand tone, and platform connections. Agents run on parallel cron schedules; account managers approve queues from one inbox.
Engagement Agent handles low-stakes interactions (likes on bookmark-worthy threads, comments matching your reply templates). You stay personal on the high-stakes threads; the agent handles the volume.
No. They handle the labor that already feels mechanical (drafting variations, scanning feeds, fetching analytics). Strategy, taste, and approval stay with you. Mirgent agents are tools that make experienced operators 5-10× faster, not replacements for judgment.
OpenAI (GPT-4o, GPT-4o-mini), Anthropic (Claude 3.5 Sonnet/Haiku, Claude 4 family), and any OpenAI-compatible endpoint. Per-agent provider selection so you can route cheap drafts through Haiku and high-stakes work through Opus.
Depends on your provider rates. Typical Content Planner run with Claude Haiku: $0.01-0.03 per draft. Lead Hunter scoring 30 leads with GPT-4o-mini: $0.05-0.10. Analytics weekly digest: ~$0.20. With BYOK and prompt caching, monthly AI bills for a single-user account are usually $5-15.
Yes (Pro plan and above). Custom agents are defined by goal + tools + cron. Wire them into workflows alongside the built-ins. Custom agents share the same memory, business-context, and approval-queue infrastructure.
Each agent run has a timeout and retry budget. After 3 failed runs the agent auto-pauses and emails you a diagnostic. Memory and goals are preserved; once you fix the cause (rate limit, expired key, etc.) one click resumes.
By default no — every public action goes through your approval queue. You can opt individual agents into auto-mode per goal (e.g. 'auto-post Content Planner drafts that score 90+ on quality check'), but the default is always approval-gated. Auto-mode is logged separately so you can audit later.
Free plan available. Pick one agent, set the cron, plug in your AI key — Mirgent does the rest.