Comments
>log in to commentHi, I'm Bryan, one of the founders of three.dev. three.dev is for developers with an AI feature in production. Paste one prompt into Claude Code, Cursor, or Codex, and it routes your LLM calls through our proxy. You keep your AI SDK. three.dev scores each request in the background and groups the failures into named failure modes, ranked by severity. Then your coding agent pulls the top one over MCP and fixes it. I'd love to hear where setup breaks for your stack.
More AI tools →74 free dev tools →
- Detect and fix agent failures in live traffic
- Score every conversation in real‑time
- Group issues into root‑cause failure modes
- Coding agent suggests and applies fixes
- Compare before‑and‑after traffic slices
three.dev scores live AI agent traffic, groups failures into modes, and helps coding agents fix them.
- for
- Teams building production AI agents who need to detect and remediate failures.
- pricing
- freemium
- license
- MIT
Key features
- Live traffic scoring — Scores every production request in real-time with no added latency.
- Failure mode grouping — Aggregates similar issues into named failure modes ranked by severity.
- Evidence-backed issues — Each failure includes the exact line or prompt that caused it.
- Coding-agent integration (MCP) — Your coding agent can query failure modes and apply fixes automatically.
- Before/after comparison — Compare metrics across two traffic slices to see impact of a fix.
- Chat interface — Ask natural-language questions about traffic and get links to underlying requests.
- Replay & experiment — Test new versions on recorded traffic before releasing to users.
Use cases
- Detect policy drift like wrong refund windows in a support chatbot.
- Identify and fix language-specific failures such as ignored non-English requests.
- Validate new prompt versions on recorded traffic before a full rollout.
three.dev pricing
- Free$0/month~50k requests/month · 1 GB data · 30 days history · Unlimited seats
- Starter$79/month~250k requests/month · 5 GB data (+$3/GB) · 90 days history · Unlimited seats
- Pro$249/month~1 M requests/month · 20 GB data (+$2/GB) · 180 days history · Unlimited seats
three.dev vs alternatives
 three.dev | .png?width=64) Langfuse |  Perfai Security |  Calljmp |  Syrin - Static Contract Analysis for MCP Servers | |
|---|---|---|---|---|---|
| Best for | Live failure detection & automated fixes for AI agents | LLM observability & analytics | Live vulnerability detection | Managed AI agent backend | MCP contract analysis |
| Pricing | Freemium | Subscription | Free | Free | Free |
| DevHunt upvotes | 1 | 89 | 30 | 35 | 24 |
| Launched | Nov 2026 | Jan 2023 | Jun 2026 | Feb 2026 | Feb 2026 |
- three.dev vs Langfuse: Langfuse provides observability for LLM apps but does not auto-fix failures.
- three.dev vs Perfai Security: Perfai Security finds live vulnerabilities, whereas three.dev focuses on agent response failures.
- three.dev vs Calljmp: Calljmp offers managed backend for AI agents, but three.dev adds scoring and failure-mode analysis.
- three.dev vs Syrin - Static Contract Analysis for MCP Servers: Syrin catches MCP server contract failures; three.dev detects runtime response failures in agents.
three.dev FAQ
What counts as processed data?+
All request and response tokens sent through three.dev; a 5k-token request is about 20 KB.
What counts as a scored request?+
Any request that three.dev evaluates; the sampling rate determines how many are scored.
What happens when I reach a limit on the Free plan?+
Your app continues to call the AI provider; recording stops after the data limit and scoring stops after the request limit, resetting each month.
How does the launch offer work?+
Sign up by Nov 22, send a request within a week, and get the Starter plan free for six months; otherwise you stay on Free.
Do you charge per seat?+
No, all plans include unlimited seats.
Summarized by DevHunt from three.dev · Nov 9, 2026. Details may change; check the official site.
