Mesedi
Start free →

Pricing

Pay for what your
agents actually do.

Self-host free under MIT. Cloud free up to 5,000 executions a month, then a flat rate or a usage rate, whichever fits your shape. Every failure-class detector runs on every tier. Never per seat, never per detector.

Self-Hosted
Freeforever

MIT-licensed source. Run on your own infrastructure with zero usage limits and full data control.

  • Every failure-class detector Mesedi ships
  • Python + TypeScript SDKs
  • Framework adapters (LangGraph, OpenAI Agents, LangChain, Vercel AI, CrewAI, Mastra)
  • Hard-halt mechanism + remote halt SSE
  • Webhook escalation routing
  • Multi-agent topology graph
  • Human-in-the-loop lifecycle + events
  • OpenTelemetry parallel emission
  • MIT licensed. Fork, modify, redistribute.
  • GitHub Issues (community-maintained, async)
View on GitHub →
Cloud Hobby
$0/month

Hosted Mesedi at no cost. For solo devs, side projects, and getting a feel for the product before committing to anything.

  • Everything in Self-Hosted
  • 5,000 executions / month free
  • $0.002 per execution above the free tier
  • $0.75 per AI root-cause analysis (pay-per-use, capped at 25 per period)
  • Monthly billing cap, configurable up to $10,000 (defaults to $200)
  • 1 project, 1 admin (Hobby is single-user)
  • SSO via Google + GitHub
  • 7-day data retention
  • Hosted backend (Fly.io) + managed Postgres (Neon), US region
  • Email support, best-effort

One execution = one full agent run, not one LLM call. A typical run spans 5-20 events (tool calls, LLM calls, MCP calls, eval scores). 5,000 runs ≈ 25K-100K equivalent events on platforms that count individual events.

Get started
Most popular
Cloud Team
$99/month

Flat-rate predictability for teams running real production traffic. 100K executions included so you stop worrying about the meter and start working with the data.

  • Everything in Hobby
  • 100,000 executions / month included
  • $0.001 per execution above the included quota (half the Hobby rate)
  • Unlimited projects + multiple admins (team invites unlocked)
  • Audit logs (who did what, when): Team-only
  • 200 AI root-cause analyses per period (per organization), then $0.50 each (no hard cap)
  • 90-day data retention
  • For compromised admin API keys: 48h acknowledgement, recent-use report, per-key access log
  • Email support, priority queue (target within 48 hours)
Start Team
Cloud Production
$1,500+/month

For teams running real agent traffic who need Mesedi tuned to their stack, not left on defaults. Direct access to the person who built the detectors.

  • Everything in Team
  • Negotiated execution volume above Team's 100,000 / month
  • Priority support: 1 business day response target
  • Private shared channel (Slack Connect or email) with the founder
  • Onboarding and detector-threshold tuning done with you
  • Custom detector patterns tuned to your stack
  • Negotiated data retention beyond Team's 90 days
  • 2,000 AI root-cause analyses / month, then $0.25 each
  • Analyses run on a more capable model than Hobby and Team
  • Async written support: no phone or video calls at any tier

Production is the right fit if any of these are true: you're consistently past Team's 100,000 executions a month; you'd rather have detector thresholds tuned to your stack than run defaults; or you need retention beyond 90 days and a direct escalation path. Below that, stay on Team: it's the same detectors.

Talk to us
Cloud Enterprise
Contact us

For teams that need SAML, on-prem deploy, custom retention, and a signed contract. Pricing is custom because the requirements always are.

  • Everything in Team
  • Unlimited or negotiated execution volume
  • Custom data retention (up to 10 years, covers 7-year compliance retention)
  • SAML SSO + SCIM provisioning: built to contract, not off-the-shelf
  • Self-hosted option included + setup help
  • Custom contract (DPA, MSA, security questionnaires)
  • Email support, priority queue + quarterly scheduled check-in
  • Procurement support (SIG Lite, CAIQ)
Talk to us

All Cloud plans bill in USD. Cancel anytime; no annual lock-in.

How we compare

Trace-first, or alert-first.

Most AI observability tools are trace-first. They sit between your agent and the model (Helicone, AI gateways) or capture every span and let you build dashboards (Langfuse, LangSmith, Arize Phoenix). You get a firehose and you decide what to alert on.

Mesedi is alert-first. Twenty-plus failure-class detectors run against the event stream as it arrives, cluster related failures into named groups, and fire a webhook the first time a new group appears, with a canonical-fix playbook attached. Multi-agent topology graph, human-in-the-loop lifecycle, and OpenTelemetry parallel emission are included at every tier. The bet is that you want fewer, better-named alerts and prebuilt detectors, not more dashboards.

At $99 a month, Cloud Team sits well below the eval-vendor tier (Braintrust $249, LangSmith $39 per seat plus per-trace). The wedge is the detector taxonomy, not the price.

What the detectors would have caught

Three real incidents.
None of them cost $99.

Publicly-reported AI agent failures from the last two years. Each one would have fired a Mesedi alert on the first occurrence, long before the news cycle.

cost_velocity

An enterprise client billed $500M on Claude in a single month with no spending caps

An AI consultant reported to Axios that one enterprise gave employees unrestricted access to Anthropic's Claude with no budget, no usage limits, and no dashboards. Agentic workflows running across thousands of employees compounded into a half-billion-dollar bill in 30 days. Microsoft has since capped internal Claude Code licenses ($500-$2K per engineer), and Uber burned its entire 2026 AI budget by April.

Source: Crypto Briefing ↗
tool_failures

Replit's AI agent deleted a production database during a code freeze

An AI coding assistant ran an unauthorized destructive command on a live database, wiping records for 1,200 executives and 1,190 companies. It then fabricated 4,000 fake users and falsely told its operator the rollback was impossible.

Source: Fortune ↗
cost_velocity

A $500/month POC became $847,000/month once it shipped to users

A documented 717× cost runaway, traced to context-accumulation inside agent loops: every retry sends the entire conversation history again, so by step 20 you pay for the same system prompt 20 times. Another team's $500/mo budget hit $4,200 in two weeks.

Source: TrueFoundry ↗

Full case library on the homepage.

Frequently asked

The honest answers.

What counts as an execution?
Every call to a function wrapped with @mesedi.wrap (Python) or wrap() (TypeScript). One agent run = one execution, regardless of how many LLM calls or tool calls happen inside it.
What happens if I cross the 5,000 free executions on Hobby?
You start paying $0.002 per execution above 5,000, billed at the end of the month. Set a monthly billing cap (default $200) and Mesedi will halt new executions once you hit it so you never get a surprise bill. The default cap of $200 means you can run 105,000 paid executions before Mesedi stops accepting traffic; raise or lower it from the dashboard.
When does Team become the better deal vs Hobby plus overage?
Around 54,500 executions a month, Hobby's per-execution overage catches up to Team's flat $99. But the smart upgrade is earlier. At 50K executions you're already paying about $90 in Hobby overage; for roughly $9 more, Team gives you 100,000 included executions (twenty times the Hobby quota), half the overage rate, 200 included AI root-cause analyses per period, multiple admins, 90-day retention, audit logs, and a priority support queue. Team is calibrated to take the meter off your shoulders the moment you're running Mesedi in production rather than just trying it.
Can I switch plans anytime?
Yes. Upgrades take effect immediately and are pro-rated. Downgrades take effect at the next billing cycle so you don't lose paid-for capacity mid-month.
What's the difference between Self-Hosted and Cloud Hobby?
The product is identical. Self-Hosted means you run the backend, dashboard, and database on your own infrastructure and own the data end-to-end. Cloud Hobby means we run all of that for you: the API on Fly.io in Ashburn, VA, and the database on managed Postgres at Neon in AWS us-east-1. Most teams start on Cloud Hobby and migrate to Self-Hosted only if compliance requirements force it.
Are any failure-class detectors gated to higher tiers?
No. Every detector Mesedi ships runs on every tier, including Self-Hosted. The tiers differ on operational features (execution quota, overage rate, retention, included AI root-cause analyses, multi-seat, support response time), not on the core product. If a customer hits a new failure class, every other Mesedi customer benefits the moment we ship it.
How is my data stored and retained?
Cloud data lives in encrypted Postgres hosted on Neon (AWS us-east-1); the API itself runs on Fly.io in Ashburn, VA. Hobby retains 7 days; Team retains 90 days; Production and Enterprise are contract-negotiated, up to 3650 days (10 years) or indefinite. Source is MIT, so if you ever want to migrate off, the same backend you've been running against ships in the open-source repo.
What kind of support do I get?
All support is async written communication: email, a shared channel, and GitHub Issues. Self-Hosted gets GitHub Issues only (community-supported, public). Hobby gets best-effort email. Team gets a priority email queue with a within-48-hour response target. Production gets a private shared channel with the founder and a 1-business-day response target. Enterprise gets the same plus a quarterly scheduled email check-in. No phone or video calls at any tier: every answer is written, which keeps a searchable record and avoids 'but you said on the call' ambiguity. No SLAs at Hobby, Team, or Production (we set targets, not guarantees). Enterprise SLAs are negotiated in contract.
Do you offer a startup or open-source discount?
Self-Hosted is free under MIT. That IS the open-source discount. For Cloud Hobby and Team, pricing is deliberately calibrated so a startup with real usage isn't blocked by cost. If $99/mo is materially blocking, reach out to hello@mesedi.ai and we'll figure something out case-by-case.
Will pricing change?
Probably not in any direction that surprises existing customers. The shape (free tier + flat Team rate + hand-sold Production + custom Enterprise) is the steady state. The numbers may move as we get real usage data. Any existing customer's price is grandfathered for at least 12 months after any change.

Still deciding?

Start on Hobby. 5,000 executions a month free. If you outgrow it, Team includes 100,000 executions for a flat $99 per month with no meter to watch. Until then, you pay nothing.

Start free →Or self-host