"48.9% of organizations are entirely blind to AI agent behavior" — Salt Security, April 2026
Annual licenses for teams running AI agents in production. ORILink stops prompt injection, unauthorized actions, and agent compromise before they happen — and gives your team full visibility into what every agent is doing, in real time, on your own infrastructure.
The tier distinction is simple: agent count. Security coverage — enforcement, ORIGuard, dashboard, audit trail — is identical across Starter, Growth, and Enterprise.
Building solo? The Individual SDK is $10 one-time. Full inbound and outbound enforcement, unlimited agents, no subscription.
Every instruction your agent receives is evaluated before your model sees it. Every action it takes is evaluated before it fires. Inbound content filtering checks where every piece of content came from and blocks anything carrying an injection signature before your agent reads it. Structured input validation catches payloads hidden in encodings and obfuscation. Outbound action enforcement evaluates what each action would actually do before it executes, and output leakage detection scans everything your agent sends for secrets that shouldn't leave. All four are unconditional. None can be disabled by operator config.
Works across GPT, Claude, Llama, Mistral, Gemma, and every other model your stack runs. Drop it in — no model changes, no framework lock-in.
ORIGuard is included in every Business tier and installs alongside the SDK. It scans your enforcement logs every 30 seconds, watching for runaway loops, credential compromise, shadow agent signatures, and cascade failures.
On a WARN event, it alerts your team and waits for a response. On a CRITICAL event — billing runaway, auth failures, injection signatures, unauthorized file writes — it suspends the flagged agent immediately, snapshots the full process state for forensics, and escalates if your team doesn't respond within 15 minutes. You investigate. The damage stops.
Alerts reach your team however they work: email, Slack, Microsoft Teams, SMS, generic webhook, or Telegram. CRITICAL alerts fire all configured channels simultaneously. WARN alerts batch to email every 15 minutes. SMS is CRITICAL-only. Everything is configurable per agent.
Every agent's token consumption is tracked in real time against a configurable daily budget. If spend hits your ceiling, ORIGuard suspends the agent before the damage accumulates. The dashboard shows session cost, daily spend, and budget progress per agent — updated every 30 seconds.
Runs locally on your machine. Five tabs: Status, Enforcement, Reports, Agents, Alerts. The Agents tab shows per-agent enforcement decisions, block rate, cost, and ORIGuard status. Reports export to JSON or CSV. Nothing is transmitted to Talonyx or any third party.
Provenance envelopes travel with every agent-to-agent message. A compromised agent can't elevate its trust weight when forwarding content to peers. Contagion stops at the first hop.
Every enforcement decision is logged with full provenance: instruction origin, trust weight, classifier result, and timestamp. Self-hosted on your infrastructure. SIEM-compatible structured export included.
Need more than 3 agents? Business Growth scales to 10. Security coverage, monitoring, and dashboard are identical — agent count is the only difference.