[FEATURE]: Decouple Kilo Memory auto-save from automatic context injection #15077

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Kilo Memory currently combines two different uses of stored memory:

1. Background capture: when project memory and auto-save are enabled, completed turns can add or update project facts and session digests.
2. Automatic context use: at the start of a session, Kilo injects a compact memory index into the prompt and can use kilo_memory_recall to retrieve additional details.

There is currently no independent control that keeps automatic capture enabled while preventing stored memory from being added to ordinary conversations. Turning memory off also disables capture, and /memory auto off disables automatic saves rather than only automatic context injection.

This makes it difficult to use Kilo Memory as a passive project archive. Stored memories may contain old decisions, invalidated assumptions, or information that is no longer relevant to the current task, so injecting them into every conversation can sometimes influence the agent's reasoning in undesirable ways.

Please add an independent user-facing control for automatic Kilo Memory context injection or recall. The exact setting name, placement, and implementation should be left to the maintainers. For illustration only, the behavior could conceptually distinguish:

{
  "memory": {
    "enabled": true,
    "auto_save": true,
    "auto_recall": false
  }
}

When automatic memory injection is disabled:

• Automatic project-memory saves and session-digest capture continue to work.
• Ordinary prompts do not receive the startup memory index or other automatic memory-derived context.
• An explicit memory request remains able to search or display stored memory.
• Existing memory management operations such as correction, deletion, rebuild, and purge remain available.
• The current behavior remains unchanged by default.
• The behavior is consistent across the CLI/TUI and VS Code surfaces that support Kilo Memory.

A concrete use case is a developer who wants every session to contribute useful project knowledge automatically, while periodically running a dedicated analysis workflow over the stored memory. That workflow could identify recurring corrections, repeated instructions, or procedures that might become rules, AGENTS.md content, or skills. Any resulting changes would still require human review. This analysis workflow is only an example consumer of explicitly consulted memory and is not part of this request.

Acceptance criteria:

• Automatic memory injection can be disabled independently of automatic memory saving.
• Automatic saves continue while automatic injection is disabled.
• No memory index or other automatic memory-derived context is added to ordinary prompts in that mode.
• Explicit memory recall and management operations continue to work.
• Existing default behavior is preserved.
• The new behavior is documented.
• Supported user interfaces expose consistent behavior.

Related:

• [FEATURE]: Keep the prompt prefix stable across turns and agent switches (follow-up to #12666) #14761 discusses prompt-prefix stability and mentions the effect of changing the pinned memory block, but does not provide this user control.
• feat: bare/minimal context mode to disable automatic prompt injection #10107 discusses a broader bare/minimal context mode; this request is limited to Kilo Memory.
• [FEATURE]: memory_model setting to run automatic memory saves on a separate model #14779 / feat(cli): add memory_model setting for automatic memory saves #14781 concern selecting a separate model for automatic memory saves, not disabling automatic memory injection.
• [FEATURE]:Feature Request: Self-Improving Agents with Selective External Memory MCP Backends #9401 concerns a much broader self-improvement and external-memory architecture.