Agentic Cache

Overview

Agentic Cache is currently a workspace settings and analytics surface. Its live bridge records hit/miss and savings outcomes from the separate semantic-cache plugin with persisted tenant, workspace, Virtual Key, kind, and optional tool dimensions. The Agentic Cache component defines generic storage methods, but no production request path calls those methods to serve a response, tool result, embedding/RAG result, or MCP discovery response in this release.

This distinction matters: a row or enabled toggle in the Agentic Cache console does not prove that an execution cache exists for that kind. Configure the live semantic-cache plugin and Virtual Key scope controls separately, then validate isolation under concurrent traffic.

Key benefits:

• Reconciled analytics - existing semantic-cache outcomes are attributed to exact-response or semantic rows with saved-token and saved-cost counters.
• Scoped reporting - persisted events are tenant/workspace scoped and can be filtered by Virtual Key. Principal class is not persisted in this event table.
• Explicit execution boundary - the console inventories exact response, semantic, tool result, embedding/RAG, and MCP discovery separately while identifying which rows are telemetry-only or reserved. Canonical GAF-governed MCP execution hard-bypasses the legacy tool-result cache.
• Persisted configuration - workspace settings and reserved controls can be saved without a restart, but only a wired execution path can consume them.

Note

Agentic Cache is available on the Business plan and above. The console lives under Workspace → Agentic Cache and reuses your Agentic Security access.

Canonical MCP results are not cached

On JSON-RPC /mcp, POST /v1/mcp/tool/execute, Agent Mode, and nested Code Mode calls protected by canonical GAF, the gateway disables both reads and writes to the legacy direct MCP result cache. Its key does not bind the complete GAF subject, delegated identity, MCP client generation, and policy decision.

The direct MCP cache and each client’s cache opt-in default to off. The Agentic Cache settings row still contains tool_result_enabled and a TTL for UI and compatibility purposes, but neither setting overrides the canonical bypass. A post-GAF, decision-bound result cache is not implemented in this release.

How it differs from the other caches

DeepIntShield exposes three related cache surfaces. Do not treat them as one shared execution cache.

Capability	What it caches	Scope
Agentic Cache (this page)	Semantic-cache outcome analytics plus reserved response, tool-result, embedding, and discovery settings	Authorization-bound reporting; it does not serve entries on production request paths
Semantic caching	LLM responses by exact + vector similarity	Per model/provider, with virtual-key scoping
Provider prompt caching	The static prefix of a prompt, at the upstream provider	Per provider

Agentic Cache attributes the savings from your semantic cache into its response (exact) and semantic kinds, so you see one reconciled $/token saved figure across the console rather than two competing numbers.

When to use it

• You want one workspace view of hit/miss and savings events emitted by the existing semantic-cache plugin.
• You need to distinguish active response-cache telemetry from reserved embedding, MCP discovery, and tool-result controls.
• You are preparing a future cache integration and want to persist intended settings now, without claiming that those settings short-circuit execution.

Configuration

All per-workspace Agentic Cache settings live on Agentic Cache → Settings. The Agentic Caches page also renders one row per modeled kind. Only the response/semantic rows receive live events, and those events describe work done by the separate semantic-cache plugin; the other rows are reserved inventory.

1. Open Workspace → Agentic Cache → Settings.

2. Under Master, turn on Agentic cache enabled to enable the analytics bridge and its workspace-level event gating. This switch does not enable the separate semantic-cache plugin or create an execution cache.

3. Under Semantic & safety, review the persisted threshold, read-only, high-risk, encryption, and obligation fields. They are reserved Agentic Cache configuration in the current runtime; they do not retune the separately configured semantic-cache plugin or enable MCP result caching.

4. Under TTLs (seconds), set the reserved Exact response, Semantic, and Tool-result values. No production Agentic Cache Get/Put call consumes these TTLs today.

5. Click Save.

6. Go to Agentic Cache → Agentic Caches to inspect the modeled kinds. Response and semantic hit rates are telemetry bridged from semantic-cache outcomes. Tool-result, embedding, and MCP discovery rows are reserved and should remain at zero unless a future release wires an execution producer. Their toggles persist intent only.

The Security Caches page lists the read-mostly caches that reduce authorization-path latency (decision/verdict, policy, key config). Cache lookup and enforcement still have measurable overhead, and misses require backing-store or policy work. The caches are invalidated structurally and on revocation push - see Agentic Security for the decision cache.

Settings reference

Field	Type	Default	Description
enabled	boolean	true	Gates Agentic Cache event recording for the workspace; it does not enable the semantic-cache plugin.
response_enabled	boolean	true	Gates exact-response outcome analytics bridged from semantic-cache.
semantic_enabled	boolean	true	Gates semantic outcome analytics bridged from semantic-cache.
tool_result_enabled	boolean	true	Reserved configuration/UI field; canonical GAF hard-bypasses direct MCP result caching.
embedding_enabled	boolean	true	Reserved configuration/UI field; no embedding/RAG execution cache is wired.
mcp_discovery_enabled	boolean	true	Reserved configuration/UI field; no tools/list execution cache is wired.
semantic_threshold	number (0–1)	0.92	Reserved Agentic Cache threshold; it does not configure the semantic-cache plugin.
semantic_read_only	boolean	true	Reserved Agentic Cache mode.
never_cache_high_risk	boolean	true	Reserved tool-result safety field; canonical GAF results remain uncached.
encrypt_at_rest	boolean	true	Reserved payload-storage field; the analytics bridge stores no cached payload.
honor_obligations	boolean	true	Reserved payload-serving field; the analytics bridge never serves a response.
response_ttl_seconds	integer	3600	Reserved TTL; not consumed by a production Agentic Cache lookup.
semantic_ttl_seconds	integer	1800	Reserved TTL; not consumed by a production Agentic Cache lookup.
tool_result_ttl_seconds	integer	600	Reserved TTL; canonical GAF tool execution does not consume it.

Cache kinds

Kind	What it reuses	Notes
Exact response	Analytics for exact hits reported by semantic-cache	Telemetry only in Agentic Cache; configure the semantic-cache plugin separately.
Semantic	Analytics for vector hits reported by semantic-cache	Telemetry only in Agentic Cache; its threshold/TTL fields do not control the plugin.
Tool result	Reserved compatibility row	Canonical GAF reads and writes are hard-bypassed; no production Agentic Cache caller serves it.
Embedding	Reserved embedding/RAG row	No production execution cache is wired.
MCP discovery	Reserved tools/list row	No production execution cache is wired.

Per-virtual-key cache scope

This section configures the live semantic-cache plugin, not the reserved Agentic Cache storage methods. Semantic reuse is bounded by the scope that distinguishes one caller from another inside a shared virtual key. By default the gateway scopes automatically; you can pin a scope mode on each virtual key for tighter control.

The available scope modes are:

Mode	A cached entry is shared across…
virtual_key	All callers using the same virtual key.
user	The same end user (resolved from request user / governance identity).
use_case	The same use_case metadata value.
session	The same session.
custom_metadata	The same value of the metadata keys you nominate.
none	No reuse - caching is effectively off for the key.

On a virtual key (Workspace → Virtual Keys → edit), the cache controls are:

• Automatic Cache - enable or disable automatic cache scoping for the key.
• Semantic Cache - enable or disable semantic matching for the key.
• Scope mode - pick the scope (virtual_key, user, use_case, session, custom_metadata, or none) and, for a custom scope, the metadata keys that define it (for example use_case, or session_id for a session-scoped key).
• Allow semantic reuse on unscoped requests - leave off if several end users share one key and you don’t want one user’s response served to another. When off, semantic lookups on a key with no per-caller scope are suppressed.
• Cache Key - an optional fixed cache key for the key. Leave it empty to use automatic scoping; a request-level x-deepintshield-cache-key header still overrides it.

Caution

In production, prefer setting a real per-caller scope (user, session, or a custom_metadata key) over enabling semantic reuse on unscoped shared keys. Leaving requests unscoped on a shared key means one caller’s response could be reused for another.

Legacy per-tool MCP cache controls

Outside the canonical GAF path, the legacy direct MCP cache can be explicitly enabled at workspace/client level and its TTL overridden per tool, including 0s to disable a tool. Each entry maps a "<server>-<tool>" name to a duration string:

{
  "search-web_search": "5m",
  "db-run_query": "0s"
}

The direct cache defaults off and only explicitly cacheable tools are eligible. It does not infer which tools are writes or high risk: operators must exclude those tools from cacheable_tools and avoid the * wildcard. These controls do not make a canonical GAF call cacheable; see MCP tool execution for the enforced boundary.

Monitoring savings

The Agentic Cache → Overview page combines several counters:

• Decision-cache hit rate from the security decision cache.
• Response and Semantic rows populated by the semantic-cache telemetry bridge. Any generic Agentic-cache hit or calls-skipped tile must be read as reported/bridged activity, not proof that Agentic Cache served a payload.
• Saved (24h) - dollars, tokens, and latency saved.
• Cross-boundary serves - a reserved tile in this release. The overview handler does not populate this counter, so its displayed zero is not evidence that a two-principal isolation test ran.

The semantic-cache plugin performs any real response short circuit. Its outcome is bridged into Agentic Cache after the response and credited with the saved tokens/cost when available. Agentic Cache itself does not perform that lookup or serve the response. Open Agent Insights → Caching for the resulting response/semantic time series; reserved kinds should not be presented as live savings sources.

Examples

Reserved Agentic Cache profile - persist conservative intent and keep MCP tool results explicitly off. This object controls Agentic Cache analytics and reserved fields; configure the live semantic-cache plugin separately:

{
  "enabled": true,
  "response_enabled": true,
  "semantic_enabled": true,
  "tool_result_enabled": false,
  "semantic_read_only": true,
  "never_cache_high_risk": true,
  "honor_obligations": true,
  "semantic_threshold": 0.92,
  "response_ttl_seconds": 3600,
  "semantic_ttl_seconds": 1800,
  "tool_result_ttl_seconds": 600
}

High-throughput chatbot with shared key, scoped per user - semantic reuse within a single end user only:

{
  "cache_enabled": true,
  "semantic_cache_enabled": true,
  "cache_scope_mode": "user",
  "cache_allow_semantic_when_unscoped": false
}

Next steps

• Semantic caching - the response-similarity cache whose savings feed the agentic cache’s response and semantic kinds.
• Provider prompt caching - cache the static prompt prefix at the upstream provider.
• Virtual keys - the boundary that scopes every cached entry, plus the per-key cache controls.
• MCP tool execution - understand canonical tool authorization and its result-cache bypass.