Which agents are running?
Your teams deploy agents independently. Nobody has a complete inventory. Shadow AI is real — and it's already in your org.
We had 12 agents in production. Oculis found 47.
Early Access · 40% off locked forever for first 50 customers
Datadog for AI agents.
Know what your AI agents cost, do, and break.
Oculis gives you full visibility into every AI agent in your organization — costs, health, performance, and relationships. One collector. Zero code changes.
No credit card required · Setup in 2 minutes · 14-day free trial
Trusted by AI teams building production-grade agents
The Problem
Flying blind is not a strategy.
Three questions your current tools can't answer — and every one of them is costing you.
Why did my AI bill spike?
$500 yesterday. $2,000 today. Which agent? Which model? Which team? You can't tell — so you can't fix it.
One bad prompt spent $3,000 overnight. We caught it on the invoice.
When did it break?
Agents fail silently. No alerts, no actionable logs. You find out when a customer complains — or the invoice arrives.
Our classifier was broken for 9 days. Nobody noticed.
If you can't see it, you can't control it. Oculis shows you everything.
How it works
From installation to insights in minutes.
One collector runs on your infrastructure. No SDK, no code changes, no manual inventory.
-
Discover
Install one collector. Find every AI agent across every machine. Zero-config detection across LangChain, CrewAI, AutoGen, LlamaIndex, Haystack, OpenClaw, or custom code.
Terminal# One line — nothing to configure docker run -d oculis/collector:latest \ -e OCULIS_TOKEN=oc_dt_... -
Analyze
Full visibility into costs, health, errors, and agent relationships. Server-side cost calculation using live provider rates — we don't trust the SDK.
- Per-agent, per-model, per-team cost breakdown
- Success rates, latency percentiles, error bursts
- Cache savings measured in dollars
-
Optimize
Get recommendations on model switches, duplicate agents, and batching opportunities. See projected savings with confidence — before you ship.
- Model optimization: “This agent would save \$420/mo on Haiku”
- Duplicate detection: two agents doing the same job
- Error-waste: tokens paid for on failed runs
-
Act
Configure alerts for cost spikes, failures, and anomalies. Set policies. Route critical incidents to the right engineer — not a team inbox.
- 12+ alert types, event-driven or scheduled
- In-app + branded email + webhooks
- Cooldowns and escalation rules prevent fatigue
From install to first insight: under 5 minutes.
Get a Demo
Capabilities
Everything you need to operate AI at enterprise scale.
Built for finance, engineering, and security teams running AI in production.
Agent Discovery
Zero-config detection across every machine. Real-time inventory with status, hardware, and topology.
- Multi-framework: LangChain, CrewAI, AutoGen, LlamaIndex, Haystack, OpenClaw
- Live topology map of agent relationships
- Host, process, and credential context per agent
Cost Intelligence
Server-side calculation, not SDK trust. 50+ models priced with live provider rates.
- Per-agent, per-model, per-team breakdowns
- Cache savings analytics measured in dollars
- Spike detection with root-cause attribution
Run Analytics
Every run, searchable. Find the prompt that cost $3,000 in minutes, not days.
- Full run history with trace-based grouping
- Error burst detection and anomaly flagging
- Latency percentiles + token efficiency per agent
Multi-Model Support
OpenAI, Anthropic, Google, Mistral, DeepSeek, Meta, Cohere. Normalized for apples-to-apples comparison.
- Full OpenRouter catalog + direct provider APIs
- Pricing snapshots refreshed daily
- A/B model switching with cost & quality deltas
Alerts & Notifications
12+ alert types. Route to the right engineer. Cooldowns prevent fatigue.
- Cost spikes, failures, anomalies, credential leaks
- Event-driven or scheduled digests
- In-app, branded email, and webhook delivery
Team & Security
Multi-user orgs with RBAC. Four separated API scopes. Row-level security at the database layer.
- JWT + Postgres RLS + app-level scoping
- Audit log for every admin action
- SOC 2, GDPR, HIPAA on the roadmap
Use Cases
Built for every stakeholder in your AI operation.
Different roles, different questions. Oculis has the answers for each.
VP of Engineering
“I need to know what my team built, how it’s performing, and where we’re wasting money.”
- Complete agent inventory across prod + dev
- Success rates & latency percentiles
- Cost optimization recommendations
- On-call integration for critical failures
CFO / Finance
“I need to control AI spend and forecast it — before the board asks me to.”
- Real-time cost tracking by team, project, agent
- Cache savings measured in dollars
- Department-level cost allocation
- Board-ready monthly summaries
Security & Compliance
“I need to know which AI is running where, with what credentials.”
- Agent inventory with host + credential context
- Credential hygiene alerts (revoked / leaked)
- Audit log exports for compliance
- SOC 2 / HIPAA / GDPR roadmap
Also built for
Why Oculis
Built for AI economics, not generic monitoring.
Every other tool was built for a different problem. Here's how Oculis is different.
vs.
Generic APM
Datadog, New Relic, Dynatrace
- Built for LLM economics — live provider rates per run, not generic CPU metrics
- Framework-aware — no custom instrumentation for LangChain, CrewAI, etc.
- Cache savings tracked as a first-class metric, not buried in logs
- Multi-provider cost normalization — OpenAI vs Anthropic vs OpenRouter, compared fairly
vs.
LLM Observability
LangSmith, Helicone, Langfuse
- Finance-first, not prompt-first — your CFO can actually read the dashboard
- Server-side cost calculation — we don't trust the SDK, we price every run ourselves
- Agent discovery — find what's running, not just log requests you already know about
- Multi-provider native — OpenRouter + direct providers in one view
vs.
Build-Your-Own
Snowflake + Looker + SDK instrumentation
- Runs out of the box — no 6-month data engineering project
- Multi-tenant SaaS or self-hosted Enterprise — your choice
- Provider rates maintained — yours will go stale in 3 months
- Security baked in — RLS, JWT, audit logs from day one
Works With
Everything you already use.
If it calls an LLM API, the collector finds it.
Frameworks
LangChain
CrewAI
AutoGen
LlamaIndex
Haystack
OpenClaw
Custom SDK
Model Providers
OpenAI
Anthropic
Google
Mistral
DeepSeek
Meta
Cohere
OpenRouter catalog
Missing one? Tell us what to support next.
Pricing
Simple, transparent, honest.
Start with what you need today. Scale when you're ready.
Core
For small teams getting started with AI.
$49
/mo
Start free trial
- 5 agents
- 10,000 runs / month
- Cost tracking & run history
- Basic alerts
- 48-hour support SLA
Most Popular
Standard
For growing teams running AI in production.
$149
/mo
Start free trial
- 20 agents up to 50 with packs
- 100,000 runs / month
- Topology maps & spike detection
- All 12+ alert types
- 24-hour support SLA
Premium
For AI-first companies and larger teams.
$499
/mo
Get a Demo
- 100 agents up to 500 with packs
- Unlimited runs
- API access & data exports
- Priority routing & escalation
- 8-hour support SLA
Coming Later This Year
Enterprise
Custom agent limits, dedicated support, self-hosted deployment, SSO/SAML, SOC 2 / HIPAA / GDPR compliance, and audit log integrations.
Ready?
Stop flying blind.
Get a demo and see why AI teams are cutting 30% of their spend without shipping less.
or
Start a 14-day free trial
Enterprise-grade security
Setup in under 5 minutes
Cancel anytime