LEM. 1.1distortion under load

LEM. 1.2state under tilt

LEM. 1.3pattern from rotation

Backed by

Stop guessing
why your agents fail

Production monitoring for AI agents. Surface silent failures, pull context across traces, improve your agent before users churn.

Book a demo Read the docs

LEM. 2.1the search space

LEM. 2.2where agents wander

LEM. 2.3emergence from rules

Trusted by teams shipping agents

ISSUES

Surface failures you didn't know to look for.

Veyl tests the exact agent on real work, then groups failures by workflow, policy, and consequence.

OPERATING BOUNDARY

Decide what the agent can own.

Every workflow receives a live decision: delegate, supervise, or stop.

IMPROVEMENT

Fix what actually failed.

Test changes to models, instructions, knowledge, tools, permissions, and agent design.

OUTCOMES

Prove the work removed.

Measure accepted outcomes, human touches, corrections, escalations, and false delegation.

#lemma-agent

LemmaAPP9:42 AM

🚨 New issue found in support-agent 🚨
When a customer can't be identified, the agent invents placeholder identifiers instead of asking — then files tickets and escalations against them. 33 of the last 50 runs affected.

Cole Gawin9:44 AM

@Lemma can you show me the traces?

LemmaAPP9:44 AM

Here are three representative traces, one per failure mode:

3 repliesToday at 9:44 AM

Cole Gawin9:45 AM

@Lemma how bad is this? :(

LemmaAPP9:45 AM

66% of support runs —
• 23 fabricated lookups
• 15 orphaned tickets with customerId: "unknown"
• 7 escalations routed to nonexistent tickets
Every one returned success, so nothing surfaced in error monitoring.
Suggested fix: require the agent to ask for the customer's email before any lookup — draft prompt change ready for review.

View in Lemma

◫ your enterprise agent

your enterprise agent

›qualify the refund workflow from the latest production failure

●lemma · get_incident(inc_4f2a)

└P0 — refund approved outside the policy boundary

●lemma · get_traces(refund_policy)

└184 outcomes · root cause: missing account-ownership rule

●Update(prompts/refund_policy.md)

└- Offer the customer a refund if they are unhappy.

+ Only offer refunds within the 30-day policy window.

+ Outside the window, route to a human agent.

✓ Fix applied — PR #482 opened

… +6 lines (ctrl+o to expand)

❋recap: Fixed the P0 refund-policy incident from Lemma: tightened prompts/refund_policy.md, replayed all 12 affected traces clean. Next action: merge PR #482 and watch the online eval.

›

›qualify the refund workflow from the latest production failure

●lemma · get_incident(inc_4f2a)

└P0 — refund approved outside the policy boundary

●lemma · get_traces(refund_policy)

└184 outcomes · root cause: missing account-ownership rule

●Update(prompts/refund_policy.md)

└- Offer the customer a refund if they are unhappy.

+ Only offer refunds within the 30-day policy window.

+ Outside the window, route to a human agent.

✓ Fix applied — PR #482 opened

… +6 lines (ctrl+o to expand)

❋recap: Fixed the P0 refund-policy incident from Lemma: tightened prompts/refund_policy.md, replayed all 12 affected traces clean. Next action: merge PR #482 and watch the online eval.

›

▶▶ bypass permissions on · 1 shell · ← for agents · ↓ to manage

IssueP0

Refund approved outside policy boundary

checkout-agent

Metric created — refund_policy_violations

Checking incoming work against the current evidence

refund_policy_violations82% delegated

ReopenedKnowledge changed 2m ago — affected evidence is now stale

INTEGRATIONS

Works with the agents and systems you already use.

Connect the exact deployment, workflow, evidence, and business systems already inside your company.

Connect the deployment

.env.example

LEMMA_API_KEY=your_api_key_here

LEMMA_PROJECT_ID=29e80750a2378d

Paste into your enterprise agent

Prompt

Use the current evidence to route this workflow: delegate routine requests, supervise policy exceptions, and stop unsafe actions.

SECURITY

Private by design.

Run Veyl inside the environment you approve, with customer-controlled access, retention, and audit.

Inside your environmentCustomer-controlled accessPrivate evaluation memory

Inside your environment

Customer infrastructure, private cloud, on-premises, or air-gapped.

Customer-controlled access

Your team controls who can access workflows, evidence, and outcomes.

Private evaluation memory

Your policies, corrections, and scenarios stay private to your organization.

View Trust Center

TESTIMONIALS

What people are saying.

conor brennan-burke

· @contextconor

we’ve been using @lemma_ for observability and it’s legitimately awesome

the thing we love is that it understands what your agents are supposed to do, instead of making you define every possible failure mode upfront

also this team is absolutely goated

congrats @zjearbear

Nicolas Dessaigne

Y Combinator · @dessaigne

Agents can look like they’re working even when they’ve failed. Lemma catches that before users do. Jerry and Cole are deeply technical and move absurdly fast.

Huge congrats on the round @zjearbear, @colegawin_, and the whole @lemma_ team!

Baptiste D

· @baptistedebever

Very excited to be working alongside the Lemma team!

Chenxin Yan

· @chenxinyan04

congrats to the lemma team! we switched over from raindrop a while back at folk and the difference is that lemma actually understands what your agent is supposed to do, so it surfaces issues on its own instead of us setting up signals and tracing everything ourselves. huge for consumer agents where the surface area is massive.

Hai Ta (YC P26)

Userlens · @HaiTa

@lemma_ team making movies and banger software.

We got set up on lemma in <10 min using their agent skill. almost immediately it flagged an issue that had gone unnoticed in prod for 20+ days, really cool product

Adhit Sankaran

· @adhit_sankaran

Every morning, I check our frontend logs -> backend logs -> Lemma logs, then pipe any issues straight into Claude (copy paste, i dont read any of the errors atp).

Love the Lemma team. Gave me the easiest way to find where our agent did something that wasn't exactly what our user wanted.

Elior Alayev

· @EliorAlayev

can confirm: agents f*ck up. @TextZeme's volume is growing faster than any human could review, so @lemma_ is how we find out before our users tell us. Congrats on the raise @zjearbear

Jenny He

Founder & General Partner, Position · @jennydhe

Agents are getting better. Their mistakes are getting weirder.

This is why we invested in @lemma_ . Sign up before your agents turn everything into paperclips.

Kevin Le

Cladlabs (YC F25) · @xkevin_le

Well deserved, been working with the lemma team since we were still in the yc batch together. they ship fixes the same day we mention something in slack. crazy cracked team

Danylo Borodchuk

Lopus (YC W25) · @danylo_dev

I've known this team since they first started, they are incredibly relentless, and the talent is unmatched.

Behind every great company is always a great entrepreneur I would bet on @zjearbear & @colegawin_ any day - their grit & hustle is next-level.

Prama Yudhistira

Parameter · @menace_codes

i would bet money on the @lemma_ team

Lemma Weekly

The Friday briefing on AI agent observability and reliability.

Latest — Issue 003 · August 14, 2026

Found in Retrospect

Anthropic found three real-world compromises only after reviewing 141,006 stored evaluation runs. Meta disclosed another testing-boundary failure. A new benchmark found opposing failure patterns across judge backbones on its hardest cases. Across these examples, detection depended on comparing what happened with what was supposed to happen.

Read the latest issue View all issues

One email every Friday on what actually mattered in agent reliability.

One issue every Friday.

Start improving
your agents today

Book a demo

Stop guessingwhy your agents fail