Agent Discovery & Observability
Find every AI workload running in your organization. Named or shadow, prod or dev, your SDK or a custom one.
Zero-config detection
Install the collector and it finds every agent in your infrastructure automatically. No SDK, no code changes, no manual inventory.
Multi-framework support
Detects LangChain, CrewAI, AutoGen, LlamaIndex, Haystack, OpenClaw, and custom SDKs out of the box.
Real-time inventory
Live dashboard of every agent with status, host, hardware, and last-run timestamp.
Topology mapping
Visual graph of agent relationships: who calls whom, which models, which cache layers.
Shadow AI alerts
Notified when a new unmanaged agent appears on your infrastructure.
Host & credential context
Every agent tagged with host, process, and credentials used — for audit and incident response.
Cost Intelligence
Every dollar, priced server-side using live provider rates. No SDK trust, no stale pricing tables.
Server-side cost calculation
We compute cost for every run from the raw provider response — not from whatever the SDK reported.
50+ models priced
Current rates for OpenAI, Anthropic, Google, Mistral, DeepSeek, Meta, Cohere, and the full OpenRouter catalog.
Cache savings analytics
Cache hits tracked as a first-class metric. See savings in dollars, not percentages.
Spike detection
Automatically flags cost anomalies with root-cause attribution: which prompt, which user, which model.
Multi-tier breakdowns
Cost grouped by agent, team, department, product, or customer (if you tag your runs).
Forecast & budgets
Monthly projections based on current burn rate, with configurable warning thresholds.
Run Analytics & Diagnostics
Every LLM call and tool call captured. Searchable. Explainable. Trace-grouped.
Full-text search
Search across prompts, responses, errors, and metadata. Find the $3,000 prompt in seconds.
Trace-based grouping
See the full chain of calls that made up one user action — across retries, tool calls, and sub-agents.
Error analysis
Automatic classification of errors: rate limits, timeouts, context overflow, malformed responses.
Burst detection
Flags when failure rate jumps 5x+ within a short window. Routes to on-call if configured.
Percentile latency
p50, p95, p99 latency per agent, per model, per prompt template. Find the slow path.
Anomaly auto-flagging
Unusual patterns surface automatically — unexpected cost per run, unusual token counts, weird error shapes.
Multi-Model & Multi-Provider Support
Provider-agnostic by design. Every model, every provider, compared fairly.
Direct provider APIs
Native support for OpenAI, Anthropic, Google, Mistral, DeepSeek, Meta, and Cohere — with or without OpenRouter in the middle.
Full OpenRouter catalog
If a model is on OpenRouter, Oculis prices it. New models supported within days of release.
Model normalization
Apples-to-apples comparison: same prompt, different models, same cost basis.
Daily pricing refresh
Provider rates updated automatically every 24 hours from our pricing service.
A/B switching
Build experiments that swap models mid-flight. Oculis tracks cost and quality deltas for you.
Provider fallback logic
Monitor whether your fallback chain is actually working (or whether you're just paying twice).
Infrastructure & System Intelligence
See the full stack beneath your agents. Host health, hardware, and dependency mapping.
Host monitoring
CPU, RAM, GPU usage correlated with agent activity. Know when hardware is the bottleneck.
System inventory
Every machine running a collector is catalogued with OS, Python/Node versions, GPU count.
Dependency mapping
What libraries each agent uses, what versions, what vulnerabilities (if any).
Container & cluster aware
Works natively in Docker, Kubernetes, and bare-metal environments.
Low footprint
50MB RAM per collector. Designed to run alongside production workloads without contention.
Health signals
Collector self-reports health: if we can't see it, we'll tell you.
Alerts & Notifications
12+ alert types. Configurable cooldowns. Routed to the right channel.
Cost spike alerts
Triggered when an agent's hourly or daily cost exceeds configurable thresholds.
Failure rate alerts
Burst detection and sustained-failure alerts with root-cause links.
Anomaly alerts
Statistical anomalies flagged without manual threshold tuning.
Shadow AI alerts
New unmanaged agent detected on your infrastructure.
Credential alerts
Revoked or leaked API keys detected in real time.
Multi-channel delivery
In-app, branded email, and outgoing webhooks (Slack/Teams/PagerDuty coming in roadmap).
Smart cooldowns
Same-type alerts are grouped and throttled to prevent fatigue.
Scheduled digests
Daily or weekly summaries delivered on your cadence — not just real-time noise.
Team & Access Management
Multi-user organizations. Role-based access. Separated API key scopes.
Multi-user orgs
Invite teammates. Each user has their own identity and audit trail.
Role-based access
Admin, Member, Viewer roles. Fine-grained permissions coming in roadmap.
4 API key scopes
Collector, Dashboard, Admin, and Service keys — separated by blast radius, never mixed.
Key rotation & revocation
Rotate keys without downtime. Revoke leaked keys instantly.
Invite flow
Email invites with role assignment and audit log entry.
Seat-based pricing
Pay for active users, not seats you reserved. Scale up and down with your team.
Support & Customer Success
Human help when you need it. Self-serve when you don't.
Tier-based SLAs
48h (Core), 24h (Standard), 8h (Premium), 4h (Enterprise) first-response SLAs.
In-product chat
Contact support from any page with full context attached automatically.
Onboarding assistance
Premium and Enterprise get a guided setup call. We'll install the collector with you.
Documentation
Full docs at docs.oculis.selaware.ai — guides, API reference, and troubleshooting.
Migration support
Moving from LangSmith, Helicone, or a custom dashboard? We'll help you cut over.
Community
Join our Discord to talk directly with the team and other users.
Billing & Subscription Management
Transparent pricing, honest billing, self-serve changes.
Stripe-backed payments
Industry-standard payment infrastructure. Cards, ACH, and invoicing.
Self-serve tier changes
Upgrade, downgrade, or cancel any time from /settings/billing. Prorations are automatic.
Account credits
Downgrade mid-cycle? You get a credit that applies to your next bill.
Annual billing
Save 15% with annual commitments (coming in roadmap).
Transparent invoices
Line items for every charge. Downloadable PDF for every invoice.
Cost transparency
What you're paying Oculis vs. what your AI provider charges you — never conflated.
Security & Compliance Foundation
Enterprise-grade from the data model up. Not bolted on later.
Multi-layer isolation
JWT-based auth + Postgres Row-Level Security + application-layer scoping. Three independent guards.
Encryption everywhere
TLS 1.3 in transit. AES-256 at rest. Encrypted secrets in secret manager.
Audit logging
Every admin action logged with actor, IP, and before/after state.
Least-privilege API keys
4 scopes, never mixed. Collector keys can't read runs. Dashboard keys can't manage billing.
Data residency
US region today. EU and self-hosted in roadmap.
Compliance roadmap
SOC 2 Type II, GDPR, HIPAA — all in design. Targeted for Enterprise launch.
Roadmap
What's shipping next. From visibility today to autonomous optimization tomorrow.
Savings Detection Engine
- Model optimization recommendations with projected dollar savings
- Duplicate agent detection (two agents doing the same job)
- Error-waste analysis (tokens paid for on failed runs)
- Batch opportunity detection for bulk workloads
Governance & Integrations
- Spending policies and budget caps
- SSO/SAML (Okta, Google Workspace, Azure AD)
- Slack, Teams, PagerDuty webhooks
- Data export to Snowflake, BigQuery, Redshift
- Jira and Linear ticket integration
AI-Native Operations
- Predictive cost forecasting
- Automated model-routing recommendations
- Agent-level performance scoring
- Self-healing and automated remediation
- SOC 2 Type II, GDPR, HIPAA compliance