1 · Inspect
Every request passes through the Altoriz edge. Managed attack signatures, virtual patches, rate limits, bot checks, threat intelligence and the LLM firewall are applied inline in under a millisecond.
The Altoriz platform
Altoriz is a web application and API protection (WAAP) platform with an autonomous AI security agent at its core. It sits in front of your websites, APIs and AI endpoints as a reverse proxy, inspects every request inline, and runs an investigation and response loop in the background.
How it works
Inline decisions take under a millisecond. The agent reasons out of band, so AI never slows a request down.
Every request passes through the Altoriz edge. Managed attack signatures, virtual patches, rate limits, bot checks, threat intelligence and the LLM firewall are applied inline in under a millisecond.
Blocked and logged requests become security events. Real-time detectors flag data leaks, bots, AI misuse and API abuse.
Every few seconds the agent groups events by source into attackers and campaigns, and analyses each: attack types, likely intent, kill-chain stage and confidence.
Under your autonomy policy the agent bans repeat attackers at the edge, deploys virtual patches and closes quiet attacks. Anything less certain stays blocked request by request.
Each action is explained in plain language with evidence, logged in the audit trail and reversible with one click.
Nothing needs you.
Altoriz handled every security event.
Illustrative example, not real data.
One clear answer on every page: what attacked you, what Altoriz did about it, and whether anything needs you. When something does get through, for example a protection that only logs, the console shows that item with exactly one fix.
Capabilities
Every capability below is available today, and every customer gets all of them. There are no tiers or add-ons.
OWASP-style protection that switches itself on, patches new CVEs at the edge and keeps sensitive data from leaving your site.
Protection against SQL injection, cross-site scripting, path traversal, command injection, SSRF, scanners and more.
The agent watches the CVE feed every 6 hours, drafts patches for technologies you run, tests them for false positives, then enforces.
Build rules by path, method, header, IP or country, or describe them in plain English and review before deploy.
Per path, per API key, per session, per country or for everyone, shared across edge nodes. Respond with block, challenge or ban.
Scoped allow rules by path, client IP range or host fix false positives without turning protection off.
Finds card numbers, personal data and secrets in responses and masks them before they leave your site.
New applications learn for 7 days, then switch to blocking automatically. Suspected false positives are held for review.
Absorbs network-layer floods at the edge, alongside Layer 7 rate limiting.
APIs change faster than rules. Altoriz learns them from clean traffic and spots the ones nobody told you about.
Learns your API endpoints from clean traffic, flags endpoints returning sensitive data, and detects shadow APIs that appear after your baseline.
Flags clients walking through object IDs to scrape or reach other users' data.
Protect expensive or sensitive endpoints such as login and search.
Enforces your uploaded OpenAPI schemas and blocks requests that don't match.
Limits query depth, complexity and batched queries.
Customers in, automated tools out, without making real people solve puzzles.
Suspected bots get a lightweight browser check. Real users pass through; automated tools don't.
Search-engine crawlers are verified, not trusted by user agent alone.
Known-bad sources and Tor exit nodes are logged or blocked by policy.
Sources that are both on a threat list and behaving like bots are raised and, under your policy, banned automatically.
Detects credential stuffing and account-takeover attempts on login flows, beyond rate limits.
Protect LLM features and AI agents from prompt injection, jailbreaks, data leakage and tool abuse, without changing your code.
Sits in front of model endpoints, detects prompt injection and jailbreak attempts, and blocks or logs them.
Redacts personal data and secrets in prompts, and caps oversized prompts.
Inspects streamed (server-sent events) AI responses.
Enforces policy on traffic between AI agents, including streaming.
Governs which tools AI agents may call on your MCP servers.
Covers the major model provider formats and their tool calls.
The agent observes, correlates, reasons, decides and acts, every few seconds, within limits you set.
Runs every few seconds: observe, correlate, reason, decide, act. Every run is recorded step by step.
Bans repeat attackers at the edge when they meet your autonomy policy, with thresholds you control.
Each attack gets a narrative, kill chain, MITRE ATT&CK mapping, evidence and recommendation.
Apply a suggested fix, for example masking data on a path, and undo it at any time.
Attacks Altoriz stopped close themselves when the source goes quiet.
All AI models run inside the platform. No traffic goes to an outside AI service.
One console for security, platform and leadership, with answers instead of alert queues.
Security score, key numbers, attack map and what needs attention.
What attacked you, what Altoriz did, and anything that needs you, on one page with filters.
A real-time request view and traffic flow map for each application.
An analysis centre with findings and threat level.
Ask questions in plain language; answers come from your live data.
PDF reports on demand or on a schedule, sent by email.
Email alerts for serious attacks, spikes, automatic bans and forwarding problems.
Architecture
A fast, stateless data plane enforces. A control plane manages policy, intelligence and the agent. The two are kept apart by design.
Inside the edge, in order
Decisions are streamed to the control plane, where the agent correlates and acts, then pushes signed policy updates back to the edge.
A stateless data plane enforces; a control plane manages policy, intelligence and the agent.
Inline decisions use compiled rules and fast checks. The agent reasons out of band.
If the edge can't load a valid policy, it keeps the last good, signed ruleset. It never silently passes traffic.
Every automatic action is logged with its reason and can be undone.
Deployment
Performance
Integrations
Forward attack summaries, blocked and detected requests, console changes and agent actions. Events are queued and retried, credentials are encrypted at rest, and mutual TLS is supported.
Elastic, the OpenTelemetry Collector and Redpanda (Kafka API) are verified end to end; other integrations are built to each vendor's published APIs. All trademarks belong to their owners.
New applications learn for 7 days before blocking, suspected false positives are held for review, and every automatic action can be undone with one click. You also set the autonomy policy.
No. All AI models run privately inside the platform. No customer traffic is sent to a third-party AI service.
Under a millisecond of inspection time (0.33 ms p99 in internal benchmarks). AI reasoning happens out of band and never slows requests.
Yes. Every security event, ban and fix has a plain-language explanation and evidence. Attacks also get a kill chain and MITRE ATT&CK mapping.
Altoriz forwards attack summaries and request events to your SIEM in its native format, and the agent reduces the noise you send there.
Yes. The data plane, and optionally the control plane, can run in your environment so payloads stay in your network.
The edge keeps enforcing the last good, signed policy. It fails closed rather than passing traffic unprotected.
Not yet. The platform is built with SOC 2 controls in mind, and certification is on our roadmap. We can share our security controls documentation. See Security and trust.
Start with one application. It learns for 7 days, then defends itself.