Private Beta

SoulGate
Deny-by-Default
Enforcement

Protect your agents' memory and MCP backends at the request layer. Zero-trust enforcement before anything reaches your data.

SoulGate is the enforcement point of Tiresias-ZT: a reverse proxy that sits between your agents and the memory / MCP backends they use. Every request is evaluated against deny-by-default policy, screened for prompt injection, metered against budgets, and logged to the tamper-evident audit trail - with no LLM in the decision path.

Request AccessRead the Docs

The problem: your agents' memory has no bouncer

Your agents read and write memory, call MCP tools, and act on what they retrieve - usually with a single shared credential and no per-request authorization. Traditional gateways rate-limit by IP and validate OAuth tokens; they have no idea what an agent is allowed to remember, retrieve, or do.

Worse, the attack surface has changed. Agents are vulnerable to prompt injection, tool-use manipulation, and automated data exfiltration. Your existing WAF does not know what an AI agent is, let alone how to protect one.

SoulGate is the enforcement point built for the agent era.

Eight capabilities for complete
request-layer security

From deny-by-default policy to injection screening. Everything your agents' backends need to stay governed at scale.

Agent-Aware Rate Limiting

Per-agent, per-tenant limits that follow the agent identity, not the IP - backed by usage metering and a cost / budget governor that stops runaway fleets.

Prompt Injection Screening

Requests are screened for prompt-injection and manipulation patterns before they reach your memory or MCP backend. Purpose-built for agent traffic.

Token Introspection

SoulAuth capability tokens introspected at the enforcement point before requests reach your backend. Expired, revoked, or malformed tokens rejected instantly.

No LLM in the Decision Path

Policy decisions are deterministic and fast - no model calls, no probabilistic allow/deny. The same request gets the same answer every time.

Sovereignty Gates

Governed, sovereignty-gated reads of memory and chain-of-thought. Content marked never-cloud is never readable from a cloud instance - by architecture, not policy.

Request Audit Logging

Every request through the proxy is logged with its decision, deny reason, and latency - SoulKey-signed, tamper-evident, and exportable.

Credential Lifecycle

Issue, rotate, suspend, and revoke agent credentials with immediate effect at the enforcement point. No orphaned keys, no zombie access.

Bring Your Own Backend

The proxy fronts the memory or MCP backend you already run. Your data stays in your store - Tiresias governs access, it never takes custody.

How SoulGate works

Every request flows through a seven-stage pipeline from ingestion to audit.

RequestInbound agent traffic
AuthToken + identity check
PolicyDeny-by-default
Rate LimitPer-key throttle
InspectInjection scanning
ProxyYour backend
AuditLog + metrics
SoulGate enforcement
Threat inspection
Upstream proxy
External

What it looks like

Front your backend, set limits, read the audit trail. (Illustrative examples - design partners get the current API reference.)

proxy.yaml (illustrative)
# Front your existing backend with the Tiresias proxy.
# The proxy runs in YOUR infrastructure - your data
# never leaves your store.
upstream:
  name: agent-memory
  base_url: https://memory.internal:8080   # your backend
policy:
  default: deny            # everything not allowed is denied
  mode: monitor            # observe first, enforce when ready
screening:
  prompt_injection: enabled
audit:
  signing: soulkey          # Ed25519 + ML-DSA hybrid
budget.yaml (illustrative)
# Metering + cost governor: runaway agents hit the
# budget wall, not your invoice.
limits:
  - target: per_agent
    requests_per_minute: 120
    action: reject
budget:
  tenant_daily_usd: 5.00
  on_exhausted: block       # fail closed, alert operators
query_audit.py (illustrative)
import requests

# Query the proxy's SoulKey-signed audit log
resp = requests.get(
    "https://tiresias.your-infra.internal/audit",
    headers={"Authorization": "Bearer <operator-token>"},
    params={"decision": "deny", "hours": 24, "limit": 5},
)

for entry in resp.json()["entries"]:
    print(entry["agent"], entry["action"], entry["resource"])
    print("  Denied:", entry["deny_reason"])
    print("  Signature verified:", entry["signature_valid"])

Completes the platform

SoulGate is the enforcement layer of the Tiresias platform. Identity (SoulAuth) + Monitoring (SoulWatch) + Enforcement (SoulGate) = complete agent security.

SoulAuth + SoulWatch provide

  • Agent identity via SoulKeys
  • Policy-based authorization decisions
  • Real-time behavioral monitoring
  • Anomaly detection and risk scoring

SoulGate adds

  • Edge enforcement before services
  • Rate limiting and circuit breakers
  • Prompt injection detection
  • Full request audit logging

Pricing

SoulGate ships as part of the Tiresias-ZT platform - it is not sold separately. Tier pricing is being finalized during the private beta.

Included with the Tiresias-ZT platform

Full SoulGate access for design partners

  • Deny-by-default policy engine
  • Prompt-injection screening
  • Usage metering + budget governor
  • SoulKey-signed audit trail
  • Sovereignty gates for never-cloud content

Put a bouncer in front of your agents' memory

SoulGate is in private beta as part of the Tiresias-ZT platform. Request a design-partner slot and start governing your backends.

Request AccessRead the Docs