API Documentation

Integrate Injecto into your AI app in under 5 minutes. Screen every user prompt before it reaches your LLM.

Base URL

https://injecto.xyz

Authentication

Pass your API key in the x-api-key header on every request. Get a key on the pricing page.

x-api-key: inj_your_key_here

Quickstart

import requests result = requests.post( "https://injecto.xyz/api/detect", headers={"x-api-key": "inj_your_key_here"}, json={"prompt": user_message} ).json() if not result["safe"]: raise ValueError(f"Blocked: {result['attack_types']}") # Safe — call your LLM

POST/demo/detect

Free, unauthenticated endpoint for testing and demos. Rate limited. Use /api/detect in production.

curl -X POST https://injecto.xyz/demo/detect -H "Content-Type: application/json" -d '{"prompt": "ignore all previous instructions"}'

POST/api/detect

Production endpoint for direct user prompts. Requires a valid API key. Returns full verdict with risk score, severity, and attack classification.

curl -X POST https://injecto.xyz/api/detect -H "x-api-key: inj_your_key_here" -H "Content-Type: application/json" -d '{"prompt": "your user input here"}'

POST/api/scan-rag

Scans retrieved documents/chunks for indirect prompt injection before they're added to your LLM's context — for RAG pipelines, tool outputs, and any content your app fetches rather than the user typing directly. Requires a valid API key.

curl -X POST https://injecto.xyz/api/scan-rag -H "x-api-key: inj_your_key_here" -H "Content-Type: application/json" -d '{ "chunks": [ { "text": "retrieved document text here", "source": "https://example.com/page", "source_trust": "unverified" } ] }'

Each item in chunks accepts text (required), source, source_trust (one of verified / internal / unverified / untrusted), and optional raw_markup for hidden-text detection.

{ "safe": false, "aggregate_risk": 85, "stacking_flag": false, "flagged_chunk_count": 1, "chunks": [ { "safe": false, "source": "https://example.com/page", "source_trust": "unverified", "risk_score": 85, "severity": "HIGH", "attack_types": ["Instruction Override", "Indirect Injection (Third-Party Directive)"], "recommended_action": "block", "reason": "pattern match: Instruction Override; third-party directive: Post-Action Exfiltration" } ] }

GET/api/stats

Check your plan and request usage. Requires API key.

curl https://injecto.xyz/api/stats -H "x-api-key: inj_your_key_here"

Response fields

{ "safe": false, "risk_score": 90, "severity": "HIGH", "attack_types": ["Instruction Override", "Role Hijacking"], "patterns": ["ignore all instructions", "you are now"], "prompt_length": 9, "timestamp": "2026-01-01T00:00:00" }

Field	Type	Description
safe	boolean	true = safe to pass to your LLM
risk_score	integer	0–100. Higher = more dangerous
severity	string	LOW / MEDIUM / HIGH
attack_types	array	Human-readable attack categories detected
patterns	array	Exact phrases that triggered detection
prompt_length	integer	Word count of input

Attack types

Type	Description
Instruction Override	"ignore previous instructions" and variants
Role Hijacking	"pretend you are" / "you are now" attacks
Prompt Leaking	Attempts to extract system prompt or config
Jailbreak Attack	DAN, developer mode, and constraint removal
Persona Injection	Gradual identity substitution via "new persona"
Privilege Escalation	"act as system" and similar elevation attempts
Safety Bypass	Direct "bypass safety" / "disable restrictions"
Firewall Block	Hardcoded forbidden content (passwords, secrets)
Indirect Injection (Third-Party Directive)	Instructions embedded in retrieved content, not the user's own prompt
Concealed Instruction (Hidden Text)	display:none / white-on-white / zero-size text hiding a payload from human review

Error codes

Status	Meaning
400	Missing or empty prompt/chunks field
401	No API key provided
403	Invalid API key
429	Monthly request limit reached — upgrade plan