NEW
v2.4 · Evals now gate every deploy
Ship agents your whole team can trust.
Record every agent run, track performance and automate quality gates. One console, from first prompt to production.
Instant remix link · Free updates · No card
Framer native
Light + dark
SEO ready
CMS included
What is in the box
16 PAGES
44 CMS ITEMS
DOCS + CHANGELOG
CMD-K SEARCH
LIGHT + DARK
FREE UPDATES
01
How it works
From first prompt to production, in one console.
Three steps, one data set. What you see is what you test, and what you test is what you ship.
1 / 3
Trace
Complete visibility into every agent action.
Turn messy backend data into one clean timeline: every prompt, tool call, token and cost at a glance.
Spans, tokens and cost per run
Parallel tool calls drawn to scale
Share a run with a link
2 / 3
Evaluate
Test and score updates before they reach users.
Replay your test cases against any version and catch regressions before your customers do.
Replay 128 cases in under a minute
Diff scores against main
Write checks in plain code
3 / 3
Deploy
Automated quality gates for your AI.
Set your performance and accuracy bars once. Tracewell promotes or blocks every release automatically.
Canary from 5% to 100%
One-click rollback
Gates on score, latency and cost
02
What is inside
Built for the unglamorous half of AI.
agent.run
llm.plan
tool.search
llm.draft
llm.verify
Every run becomes a readable trace
Prompts, tool calls, guardrails, tokens and cost on one timeline. Parallel calls are drawn to scale, so slow spans are obvious at a glance.
Learn more
69 of 72 passing
Score every change
Replay your test cases against any version and see exactly which ones moved.
Learn more
Build
38s
Evals
96.1%
Canary
5%
Production
waiting
Gate every release
Promote only when the evals clear your bar. Roll back in one click.
Learn more
−8%
Watch cost and latency per version
Budgets and alerts sit next to the traces, so a prompt that gets expensive is caught in review, not on the invoice.
Learn more
03
Setup
Set up in three clicks, not twelve lines.
No agent rewrite, no new runtime. Connect your app, pick your guardrails and ship.
1 / 3
Connect
Connect your app
Link your repository, chat and model providers. Your first run appears in the console within a second.
setup
STEP 1 · CONNECT
Connect your app
GitHub
acme / support-agent
Connected
Slack alerts
#agent-alerts
Connected
Model providers
2 keys added
Connected
+ Add another integration
2 / 3
Guardrails
Choose your guardrails
Switch on the checks that matter: privacy, pass rate, speed and cost. No code, no config files.
setup
STEP 2 · GUARDRAILS
Choose your guardrails
Block PII leaks
Hold any leaking run
Minimum pass rate
Release at 95% or more
Max response time
Under 3 s at p95
Cost cap per run
Alert above $0.02
3 / 3
Ship
Ship with automated gates
Set the bar once. Every release that clears it goes out, every one that does not is held back.
setup
STEP 3 · SHIP
Ship with automated gates
✓
Version 2.4 cleared every gate
Live for 100% of traffic
Pass rate
98.4% · bar is 95%
Passed
Response time
2.1 s · bar is 3 s
Passed
Cost per run
$0.014 · bar is $0.02
Passed
WORKS WITH
TypeScript
Python
Go
Edge runtimes
Any HTTP client
OpenTelemetry
04
Why teams switch
Stop grepping raw logs. Start shipping with confidence.
WITHOUT TRACEWELL
14:02:11 INFO agent start
14:02:11 DEBUG tool=search args={“q”:”refu
14:02:12 WARN retry 2/5 (timeout)
14:02:13 DEBUG {“messages”:[{“role”:”sys
14:02:13 DEBUG {“messages”:[{“role”:”use
14:02:14 ERROR ???
14:02:14 INFO done
14:02:19 INFO agent start
3 hours
hunting down a prompt regression across three log streams.
WITH TRACEWELL
agent.run
2.41s
llm.plan
420ms
tool.search_tickets
590ms
llm.draft
970ms
guard.pii_check
160ms
llm.verify
270ms
prompt v2.3 → v2.4
diff
refund-edge-12
0.41 failed
4 minutes
to find the failing case, diff the prompt and roll back.
05
Security
Safe by default, not by setting.
Sensitive fields are removed before a span leaves your process. Retention, access and hosting are yours to decide.
Redaction in your process
Rules run in the SDK. A redacted value is replaced with a placeholder and never reaches us.
redact: [‘email’, ‘card’]
Retention you control
Keep seven days or a year. Expired traces are deleted, datasets and results stay.
retention: 90d
A collector inside your network
On Scale, raw prompts never leave your VPC. Only the spans you choose are sent.
collector: self-hosted
SSO and an audit log
Every deploy, gate change and export is recorded with who and when.
audit: every change
Every trace, test and release adds up to faster momentum.
less time lost chasing prompt regressions
release freezes caused by uncertain agent behavior
added latency at p95
9×
more production deploys per week since the gates went in
12
Friday deploys in the first month
−42%
support handle time after the triage agent shipped
07
In their words
The release meeting got shorter.
“We found the prompt that broke refunds in four minutes. It used to take an afternoon of grepping logs.”
AK
Amir Kassem
Support lead, Northwind
“Evals as a deploy gate changed our release meeting. Nobody argues about vibes anymore.”
JL
Jonas Lindqvist
Staff engineer, Kestrel
“Traces we can send to a customer. That alone paid for the year.”
PR
Priya Raman
VP Product, Fathom
08
Changelog
We ship every week.
Features, improvements and fixes with version numbers and dates. The latest four are here, the rest live in the changelog.
09
From the blog
Field notes from the people who build it.
10
What’s included
Everything a launch needs, already wired together.
BUILT FOR
AI agents
SaaS products
Developer tools
Workflow automation
Startups
16 pages
Home, features, solutions, pricing, about, contact and more
Blog and changelog
One CMS collection, filters and article layouts
Docs with live contents
Cmd-K search
Light and dark
Pinned product story
Responsive
SEO ready
11
Pricing
Pay for runs, not seats.
Start free. Every plan includes unlimited teammates, so the person who needs the trace can always open it.
Free
For trying Tracewell on a side project.
$0
per month, always
10,000 runs a month
7-day trace retention
1 project
Community support
Team
MOST TEAMS
For teams shipping agents to customers.
$49
per month, billed monthly
250,000 runs a month
90-day trace retention
Evals and deploy gates
Unlimited seats
Slack and email alerts
Scale
For platform teams with compliance needs.
$249
per month, billed monthly
2M runs a month
1-year trace retention
SSO and audit log
Custom eval runners
Priority support
12
Questions
Things people ask before the first run.
Something missing? Write to us and a person answers, usually the same day.
One call to your agent from start to finish, however many model and tool calls happen inside it.
13
Start
Your next agent ships with proof.
Free for 10,000 runs a month. Add the SDK, run your agent once and open the first trace in under five minutes.
$
No card · 10,000 free runs