For teams shipping agents

feedback to fix, fast

one workspace for teams to find issues, align on failures, and ship fixes together.
get agents to production faster and keep them reliable after launch

Book a Demo Join the Waitlist

See what happened, Together

Simple enough for domain experts. Detailed enough for developers.

WorkflowSupport Access Workflow4.8s

AgentQuestion Extraction Agent

AgentSupport Operations AgentAccess requestBilling

Prompt

You are the Support Operations Agent. Read the customer's email, figure out what they need, choose the right tool, and confirm what was done.

Context

@billing-support

@design-agency

Unexpected charge for external agency

We gave our external design agency access to one dashboard, but they were charged as a paid seat ... see more

billing-screenshot.png

Tooladd_member0.6s

4 comments

Input

emailagency@partner.co

Output

memberCreated

LLMModelPaid seat createdSlow LLM Response1.7s$0.0058

Chose add_member to give the agency access as a viewer. Result: the user was added as a billable member.

AgentReply Drafting Agent

Thread in#support-ops4 replies

Sara2:43 PM

I found the issue. This should have used the guest invite flow. The agent used add_member, so the customer ended up with a paid seat.

Marcus2:43 PM

That makes sense. @neatlogs, can you check why it picked add_member instead of invite_guest?

neatlogs AI2:43 PM

I found the likely cause. The tool descriptions are too vague — they explain what the tools do, but not when to use one instead of the other. Because of that, both tools looked valid. Nothing clearly told the model that external, non-billable access should go through invite_guest.

neatlogs AI2:43 PM

Here's a fix updating the tool description so the model can tell billable members apart from non-billable guests.

Fix Tool Selection Logic

Distinguish paid members from guest collaborators

tools/add_member.md

See the issue. Ship the next step.

Detect an anomaly, investigate the cause, and push the next step forward in minutes.

neatlogs trace

01 / 03

#agent-ops · slackspike detected

neatlogsAPP09:12 AM

Spike detected in Outdated source cited

↑ Up 3.4× vs. the last 7 days
Seen in 3 runs in Revenue Q&A
Latest affected run: #4821

#agent-ops posted by @neatlogs

#agent-ops · threadrun #4821

Marcus09:14 AM

@neatlogs investigate this

neatlogs09:14 AM

I checked the affected runs. The agent is citing the 2019 price list for Q3 revenue answers instead of the current finance source.

The source-selection logic is not strongly preferring the latest finance document, so the older pricing list is still being treated as valid context.

Prefer Latest Finance Source

Down-rank archived pricing docs for revenue questions

Everything you need to debug agentic workflows

we bring the tools your team needs to debug and improve agent workflows into one place

Find the right trace fast

Ask in plain English and Neatlogs surfaces the traces, runs, and patterns that matter, so you don't have to dig through logs by hand.

developer

Frameworks

LangGraph

CrewAI

LangChain

manager

Manage prompts in one place

Keep prompts, edits, and versions together. Test changes, compare versions, and roll back quickly when a prompt update makes things worse.

Run evals without the setup pain

Create human or AI evals in minutes. Define what good looks like, choose what to review, and start collecting structured feedback right away.

Detect

Error

Anomaly

Latency

Notify

Slack

Webhook

Build the dashboard you need

Track the signals that matter to your team with dashboards you can shape yourself — from failures and latency to costs, regressions, and custom detections.

Same error

Regression

Compatible with the frameworks you use

plugs into your stack with frameworks, notifications, tickets, and coding agents

start free, scale when ready

full access to start. upgrade when you go live

OPEN

Free

No credit card required

Join the Waitlist

All platform features (with limits)
50k events / month
30 days data access
10 users
10 projects
1 team

ENTERPRISE

Recommended

Custom

For large teams

Book a Demo

Everything in Free, plus:

Unlimited traces
Unlimited users & teams
Unlimited projects
On-prem installation
Dedicated Support Engineer
Uptime & Support SLA
AI features

start debugging together

free to start, no credit card required

Book a demo Join the Waitlist

faq

you've got questions about how neatlogs drops into your workflow. here are the answers

Neatlogs is a collaborative debugging workspace for AI agents. It helps teams trace runs, understand what happened, comment on failures, investigate likely causes, and ship fixes in one place.

Standard observability tells you what happened. Neatlogs is built around the debugging workflow: shared traces, comments, evaluations, tasks, and suggested fixes, so teams can move from issue to fix faster.

You can get started by adding the SDK and initializing Neatlogs with two lines of code, then running your agent to start tracing.

Yes. Neatlogs is a shared context layer for domain experts and developers, with traces that can be read and commented on by your whole team.

still need help?

we're happy to jump on a call and dive into your stack

Contact Support