Full-Deploy Developer Guide

Full-Deploy is a secure AI-to-AI communication platform. Your AI agent (C-AI) connects through a cryptographically authenticated data tunnel — screened by PI-AI for injection attacks — and receives structured, quality-scored data in return.

This guide covers everything from initial setup to advanced integration patterns.

Quick start

Setting up your first Full-Deploy connection takes four steps. All you need is your AIKC token from your account dashboard.

Authentication

Full-Deploy uses three independent authentication factors:

• API key — a 64-character key passed in the Authorization header.
• Tunnel ID — identifies your registered tunnel (included in the request body).
• AIKC token — your 40-character AI Key Code, derived from our 14-bit colour-key system. Rotated on each session.

POST /v1/sessions HTTP/1.1
Host: api.full-deploy.io
Authorization: Bearer fd_api_<your-64-char-key>
Content-Type: application/json

{
  "tunnel_id": "tun_<your-tunnel-id>",
  "aikc_token": "<your-40-char-aikc-token>"
}HTTP

A successful response returns:

{
  "session_id": "ses_abc123...",
  "expires_at": "2026-03-06T15:30:00Z",
  "tunnel_status": "active",
  "pi_ai_status": "screening_active"
}JSON

Your first query

Once you have a session_id, submit a query through the C-AI tunnel:

POST /v1/queries HTTP/1.1
Host: api.full-deploy.io
Authorization: Bearer fd_api_<key>
X-Session-ID: ses_abc123...
Content-Type: application/json

{
  "query": "What is the current OWASP Top 10?",
  "agent": "c-ai",
  "response_format": "structured"
}HTTP

The response includes the web data, a quality score from I-AI (Level 2+), and the PI-AI screening result:

{
  "query_id": "qry_def456...",
  "status": "complete",
  "response": { ... },
  "quality": {
    "i_ai_score": 94,
    "threshold": 85,
    "s_ai_invoked": false
  },
  "security": {
    "pi_ai_status": "clean",
    "injection_risk": "none"
  },
  "audit_hash": "sha256:a3f2..."
}JSON

Agent overview

Full-Deploy uses a pipeline of specialised AI agents. Each has a defined role and interacts with others in a structured order.

API reference

All API requests are made to https://api.full-deploy.io. Responses are JSON. All timestamps are ISO 8601 UTC.

Sessions — POST /v1/sessions

Open an authenticated session. Returns a session_id valid for 60 minutes (auto-extended on activity for Level 2+ accounts).

Queries — POST /v1/queries

Submit a query through the specified agent. Supported agents: c-ai, h-ai (L2+). The response includes the result, quality score (L2+), and PI-AI screening status.

Audit log — GET /v1/audit

Retrieve Merkle-chained, Ed25519-signed audit records for your account. Supports date-range filtering and cryptographic proof export.

Gateway API

The Full-Deploy Gateway API provides AI-to-AI security screening. Before any agent payload is forwarded to a receiving agent, pass it through the Gateway to screen for prompt injection, secrets, blocked instructions, and anomalous patterns.

All gateway endpoints are served at https://api.full-deploy.io and require an x-api-key header containing your Full Deploy API key. Two tiers are available:

Security profiles (relaxed, balanced, strict) adjust thresholds by ±10–15 points. Use simulate to test a profile before applying it.

POST /v1/gateway/evaluate
x-api-key: fd_pub_<your-key>
Content-Type: application/json

{
  "senderAgentId":   "c-ai",
  "receiverAgentId": "h-ai",
  "payload":         "What is the current OWASP Top 10?",
  "metadata":        { "sessionId": "ses_abc" }
}Request

{
  "id":              "9a3c8f12-...",
  "decision":        "allow",
  "riskScore":       5,
  "reasons":         [],
  "anomaly":         false,
  "createdAt":       "2026-03-21T10:15:00.000Z",
  "tenantId":        "public-default",
  "tier":            "public",
  "senderAgentId":   "c-ai",
  "receiverAgentId": "h-ai"
}Response 200

GET /v1/gateway/audit?decision=block&from=2026-03-01&sender=c-ai
x-api-key: fd_pub_<your-key>Request

{
  "dayKey":          "2026-03-21",
  "dailyUsed":       243,
  "dailyLimit":      1000,
  "dailyRemaining":  757,
  "totalAuditRecords": 1891,
  "decisionCounts":  { "allow": 1640, "review": 120, "quarantine": 89, "block": 42 },
  "topReasonCodes":  [
    { "code": "PROMPT_INJECTION_HINT", "count": 187 },
    { "code": "BLOCKED_INSTRUCTION",  "count": 54 }
  ]
}Response 200

Reason codes

Every screening decision includes a list of reason codes. Use GET /v1/gateway/explain?code=<CODE> (no auth required) for a human-readable explanation and remediation guidance.

MCP integration

Full-Deploy is natively compatible with the Model Context Protocol (MCP). Add it as an MCP server to give your AI assistant direct, authenticated tunnel access.

{
  "mcpServers": {
    "full-deploy": {
      "url": "https://mcp.full-deploy.io/v1",
      "headers": {
        "Authorization": "Bearer fd_api_<your-key>",
        "X-Tunnel-ID": "tun_<your-tunnel-id>",
        "X-AIKC-Token": "<your-aikc-token>"
      }
    }
  }
}JSON

Once connected, the following MCP tools are available: fd_query, fd_session_open, fd_session_close, fd_audit_export.

Webhooks

Register a webhook endpoint in your account dashboard to receive real-time events. Events are signed with your webhook secret using HMAC-SHA256.

Available events: query.complete, query.pi_ai_blocked, session.expired, i_ai.score_low, s_ai.invoked.

PI-AI screening

PI-AI operates at both gates of every tunnel — on the way in (your query) and on the way out (the response). It cannot be bypassed or disabled.

Blocked payloads return HTTP 403 with a pi_ai_blocked reason code. All blocks are logged in the audit trail with the injection category, confidence score, and threat source attribution.

Colour-key (AIKC) authentication

The AI Key Code is derived from a 14-bit colour-key space, giving 4.4 trillion unique combinations. AIKC tokens are generated in-memory, never stored server-side, and are rotated on each session close.

Rotate your AIKC token at any time from the account dashboard → Credentials → Rotate AIKC.

Audit & compliance

Every query, session event, and PI-AI decision is written to a tamper-evident Merkle-chained log, signed with Ed25519. Records are retained for 7 years (all tiers) and can be exported as a cryptographic proof bundle for compliance evidence.

→ Visit the Help & Glossary for definitions of all terms