Optic
See if your AI features work, per customer
Last verified:
What is Optic?
Optic is an open-core LLM observability platform designed for enterprise teams who need visibility into AI feature quality, cost, and agent execution while keeping data inside their own infrastructure. It provides two-line SDK integration to track every AI interaction, capturing input fingerprints, output quality signals, latency, and user behavior without storing raw prompts or outputs.
The platform offers two specialized views: an Engineering view that shows feature quality and regressions within hours with real-time alerts for quality threshold breaches, abandon rate spikes, and latency degradation; and a CS view that provides per-account health scores, silent struggle flags, and daily QBR briefs showing which accounts had AI struggles overnight and which renewals need attention.
Optic is built for SaaS teams shipping AI features who want to know if they're actually working. It detects quality regressions before support volume spikes, tracks cost drift with per-call spend and budget alerts, makes agent opacity visible through span-level debugging, and includes built-in compliance workflows for data export, retention policies, and audit logs.
Key features include automatic quality scoring for every call, built-in hallucination detection for low-confidence outputs and refusal loops, an agent waterfall debugger for multi-step run tracing, cost observability with per-model spend breakdowns, window-based regression monitoring for quality drift detection, and compliance workflows with GDPR export and retention flows.
Optic pricing
Pricing model: Freemium
Free tier: EUR 0/month - Ingest API, Basic stats (24h), API-key project isolation, Prompt Debugger preview (demo, read-only). Starter: EUR 299/month with 12-month term billed monthly - Everything in Free plus Prompt Debugger, Managed hosting. Team: EUR 999/month with 12-month term - Everything in Starter plus Agent Debugger, Advanced analytics, Compliance readiness preview. Scale: EUR 2,999/month with 12-month term - Everything in Team plus Compliance/GDPR workflows, Governance-ready controls. Enterprise: EUR 5,000+/month - Everything in Scale plus Private deploy, SLA and security package. Prices exclude VAT. Standard checkout supports card payments (Apple Pay, Google Pay) and SEPA direct debit.
Optic pros
- Two-line SDK integration with OpenAI, Anthropic, or custom models
- Account-level visibility by design, not aggregated averages
- Real-time engineering alerts within hours of quality issues
- Daily CS digest with accounts needing action ranked by urgency
- No raw prompts or outputs stored, only quality signals
- Automatic quality scoring without manual eval setup
- Built-in hallucination detection for low-confidence outputs
- Agent waterfall debugger with per-span timing and cost visibility
- Per-model and per-feature spend breakdowns with alert thresholds
- Window-based quality drift detection for early warning
- Built-in compliance workflows with GDPR export and retention
- Open-core platform with public core plus enterprise modules
- Managed hosting option with SLA and support for paid plans
- Works in production without proxy in critical path
- Free tier available with no credit card required
- Audit-ready data flows for security and legal stakeholders
- Silent AI struggle detection before customers churn
Optic cons
- Paid plans require 12-month terms billed monthly
- Prices exclude VAT, adding cost for EU customers
- Free tier limited to 24h basic stats only
- Prompt Debugger only available on Starter plan and above
- Agent Debugger only on Team plan (EUR 999/month) and above
- Compliance/GDPR workflows require Scale plan (EUR 2,999/month)
- Enterprise plan pricing starts at EUR 5,000+ with custom quote
- Managed hosting only on paid plans, self-hosted is evaluation-only
Frequently asked questions about Optic
Why pay if a public core repository exists?
The public core is for evaluation and integration speed. Paid plans are for reliable production operations, governance, and enterprise accountability. Paid plans include managed operation with production hosting and upgrades, SLA and support with defined uptime targets and response times, security package with enterprise trust artifacts, and plan-gated capabilities like Prompt Debugger, Agent Debugger, advanced analytics, and compliance workflows enforced server-side.
How does AgentLens differ from product analytics tools like Amplitude or Mixpanel?
Product analytics tools track clicks and funnels. When a user clicks an AI feature and gets bad output, they log one event (a click). The abandon, frustration, and silent decision to stop trusting it remains invisible. AgentLens captures every AI interaction with input fingerprint, output quality signals, latency, and what the user did next, providing account-level visibility by design since the customer who churns isn't in your averages.
How does AgentLens differ from observability tools like Datadog or New Relic?
Observability tools monitor infrastructure and uptime. They can show everything is green while 54% of AI feature interactions end in abandonment. Uptime is not quality. AgentLens specifically tracks AI feature quality, output degradation, abandon rates, and silent struggle signals that infrastructure monitors miss.
How does AgentLens differ from CS platforms like Gainsight or ChurnZero?
CS platforms track logins, NPS, and support tickets. A customer can log in every day for weeks after silently giving up on AI features while their health score shows green. They may renew in 31 days without ever raising a ticket. AgentLens detects silent AI struggle at the account level before customers churn, showing which accounts had AI struggles overnight and which renewals need attention.
How long does integration take?
Integration takes about 10 minutes with one npm package. You wrap your AI calls and it works with OpenAI, Anthropic, or any custom model. The integration is just two lines of code using import agentlens, agentlens.init(), and agentlens.patch_openai().
What data does AgentLens store?
AgentLens captures every AI interaction in real-time including input fingerprint, output quality signals, latency, and what the user did next. No raw prompts or outputs are stored, only quality signals. This keeps application data inside your own infrastructure while providing visibility into quality, cost, and agent execution.
What alerting does AgentLens provide for engineering teams?
Engineering gets real-time alerts within hours of quality issues. Quality threshold breaches, abandon rate spikes, and latency degradation surface the moment they happen. For example, when a prompt update collapses AI feature quality, AgentLens would fire an alert within 2 hours, not weeks later when a customer complains.
What does the CS team see each morning?
Every morning, the CS team sees exactly which accounts had AI struggles overnight, which health scores changed, and which renewals need attention that day. No manual report pulling or Salesforce queries required. Just the accounts that need action, ranked by urgency, in a daily digest.
Can I try AgentLens before committing to a paid plan?
Yes. The Free tier is available with no credit card required and includes Ingest API, Basic stats (24h), API-key project isolation, and Prompt Debugger preview (demo, read-only). You can also open the dashboard, traces, and compliance preview with demo data first, then request a guided 14-day pilot.
What payment methods does AgentLens accept?
Standard checkout supports card payments including Apple Pay and Google Pay where available, and SEPA direct debit. Starter, Team, and Scale plans use 12-month terms billed monthly. Prices exclude VAT.