Mirgent never marks up AI usage. Connect your own OpenAI, Anthropic, or other provider key; we orchestrate the calls. Costs land directly on your provider bill, fully transparent. No opaque credits, no surprise overages, no AI-as-loss-leader pricing games.
Six reasons opaque-credit AI pricing is bad for everyone except the vendor.
Mirgent doesn't resell tokens. Whatever your provider charges, that's what you pay. We make money on the platform; AI is your relationship, not ours to monetize.
OpenAI (GPT-4o, GPT-4o-mini, GPT-4 Turbo), Anthropic (Claude 3.5 Sonnet, Haiku, Claude 4 family), and any OpenAI-compatible endpoint. Per-agent provider selection.
Agency users: each client workspace can use a different AI key. Client A's costs land on their provider account; you never co-mingle billing.
Hard caps prevent runaway spend. Worst-case is one agent's daily budget exceeded, never a $4,000 surprise from an infinite-loop bug.
Mirgent structures every prompt to maximize provider-side cache hits. On supported models (Claude, GPT-4o), cache hits cut token costs 40-90%.
Real-time per-agent token usage and provider-side cost surfaced in Mirgent. Spot expensive agents; tune token budgets; switch to cheaper models per agent.
One-time setup. After that, Mirgent uses your key for every AI call until you swap providers.
OpenAI: platform.openai.com → API Keys → Create. Anthropic: console.anthropic.com → API Keys. Other providers: same pattern, takes 1-2 minutes.
Settings → AI Providers → Add Key → paste. Mirgent encrypts at rest with per-workspace keys; never visible in logs.
Set the default for your workspace (e.g. Claude Haiku for cheap drafts, GPT-4o for high-stakes). Per-agent override available.
Per-agent hard cap on daily token consumption. Recommended: 100K tokens/day for Content Planner, 50K for Lead Hunter. Cheap default; raise as needed.
First week: monitor spend daily. Most users land at $5-15/month for solo / $50-150/month for agencies. Anomalies flag in red on the dashboard.
Mix-and-match per agent. No reason to lock into one provider.
$0.25/M input tokens, $1.25/M output. Best for short-form drafts (Twitter, captions, comments). Quality:cost ratio unmatched for sub-300-token outputs.
$3/M input, $15/M output. Best for medium/long-form (LinkedIn long-form, blog summaries, lead replies). Voice quality is the strongest among providers.
$0.15/M input, $0.60/M output. Cheapest mainstream model. Use for lead scoring and bulk classification where token cost dominates.
$2.50/M input, $10/M output. Use for technical / code-heavy drafts (Hashnode, Dev.to). Stronger than Claude on code-block formatting.
Premium tier. Use for high-stakes long-form (annual reports, founder essays). 10-100× quality lift over Haiku at 100× the cost.
Groq, Together AI, Fireworks, self-hosted vLLM, anything with an OpenAI-compatible endpoint works. Use for cost arbitrage on bulk classification jobs.
Three honest profiles from Mirgent's user base. BYOK transparency means we know exactly what people pay.
A handful of AI drafts/day, mostly Claude Haiku, across a few platforms. Monthly AI bill: $1-3 on your own provider key. Mirgent Pro: $19/mo (AI generation, agents and workflows are paid-plan features). Transparent, pass-through AI cost.
30 lead drafts/day + content planner running daily, mostly Sonnet. Monthly AI bill: $15-30. Mirgent: $19/mo. Total: ~$40/mo to replace a $3K/mo marketing hire.
20 clients, each with their own key (cost passed through). Each client's monthly AI: $5-15 (their bill). Agency: $49/mo flat. Margin stays clean.
API keys are sensitive credentials. Mirgent treats them like passwords.
Three reasons: (1) AI prices change weekly; we'd have to constantly re-price. (2) Different users want different models for different tasks; one-size-fits-all is a worse product. (3) Markup-on-AI is a bad business model long-term, eventually customers find out and feel cheated.
Same as if you ran out of credits using the API directly: the call fails with a 429 / quota error. Mirgent surfaces this in the dashboard immediately and pauses agents until you top up. Nothing breaks silently.
Yes, agent configuration is provider-agnostic. Switch from OpenAI to Anthropic with one click; same agent, different model. Useful for cost optimization or when one provider has an outage.
OpenAI and Anthropic both let you sign up with $5-10 credit minimum. For most Mirgent users, $5 of credit lasts a month. Linking a credit card later is optional.
Yes, we have to, in order to construct prompts and process outputs. Prompts are not used to train any model. Logs are kept 30 days for debugging then purged. If your organization has stricter compliance needs, reach out and we'll work with you.
Absolutely. AI is optional everywhere, Mirgent is a full scheduler + lead-discovery + workflow tool without AI. Many users start AI-free, validate the workflow, then add a provider key once they see clear ROI.
Free plan supports BYOK. $5 of OpenAI credit lasts most users a month, see real costs, no markup.