A firewall and antivirus for autonomous agents. D.A.T.D.A scans every prompt, tool call, memory write and inter-agent message — and blocks prompt injection and the other dark arts before they execute.
Integrate in five minutes. Cheapest-integration first.
POST /v1/scan with the text or tool call. Get a verdict — allow, flag, or block — in one round trip.
Add one endpoint. Your agent gains scan and check_tool tools in Claude, Cursor, or any MCP host.
Autonomous agents discover, self-provision, and buy scan credits within a spend cap — no human in the loop.
Mapped to OWASP Agentic Security (ASI01–10) and NIST AI 100-2.
| Channel | What it stops | OWASP |
|---|---|---|
input | Direct / indirect / obfuscated prompt injection | ASI01 |
tool | Unauthorized tool calls, dangerous params, exfil URLs | ASI02/06 |
memory | Poisoning of MEMORY.md, hooks, shell config | ASI04 |
inter_agent | Spoofed / injected messages between agents | ASI07 |
goal | Intent drift & goal override across a sequence | ASI03 |
The signatures are table stakes. The differentiation is a real-time, streaming causal knowledge graph of attacks — running in our private core, reached only through the API.
Start free. Pay as you scan. Agents buy their own credits.
10,000 scans a month, no card required.