Your AI agents are making decisions you can't see. One bad decision can cost days of debugging or production downtime. Fae shows you exactly what happened, token-by-token.
Missing context, swallowed state, unreproducible errors: without a complete record, you're left asking "why did it do that?" while combing logs and rerunning jobs.
One bad token in a prompt chain can derail everything.
Without a state trail, bugs slip through and resurface in production.
You need observability that's lightweight and language-native.
Fae records every prompt, response, and memory change your agents make, so you can replay, inspect diffs, and fix issues in minutes instead of hours.
See every prompt, response, and memory diff in sequence.
Pinpoint exactly where things went wrong.
Drop-in SDKs for Python & Ruby. No agent logic changes.
Share traces, annotate steps, keep everyone aligned.
from fae import Fae
fae = Fae(api_key="YOUR_KEY")
fae.observe(your_agent)One line to install. Capture your first run in under 5 minutes.
Start free with generous limits, then scale as you grow. No surprises, no hidden fees.
Perfect for trying out Fae with your side projects
For teams serious about debugging AI agents
Lock in launch pricing forever
Need more? Enterprise plans with custom limits, SLAs, and dedicated support. Contact sales
Idiomatic Python & Ruby.
Works with any framework or loop.
No guesswork needed.
Minimal overhead, no vendor lock.
Timeline, diffs, search without bloat.
100 free steps/day → 50k+ with same SDK.
Install the SDK and add a one-liner. No architecture changes needed. Most teams capture their first run in under 10 minutes.
Fae is observer-only and asynchronous. Typical overhead is less than 1% for agent response times.
Yes. Data is encrypted in transit and at rest. Enterprise plans include SSO and audit logs.
Yes. Fae is agent-agnostic and framework-agnostic. Works with OpenAI, LangChain, custom loops, and more.
Free up to 100 steps/day. Pro is $149 (this week only! Regular price: $249) for 50k steps/month. Team and Enterprise plans coming soon.
On-prem or private cloud options are available on Enterprise plans.
Join 2,000+ teams who've cut debugging time by up to 73%.