dsh-missher-brain

In a DeepSeek Harness profile where Memory and MSE each own their data and privacy rules, the agent still needs a single bounded recall entry point. dsh-missher-brain exists for exactly that: it is a thin integration layer that does not create a database, train a model, or add another panel. It injects registered-provider recall into the first step of every top-level turn that contains direct user text.

On the capability side, Brain supplies the missherBrain service and the ordinary Settings → Memory & Learning panel. Every recall is strictly bounded—at most 6 items and 4,000 UTF-8 bytes including the safety wrapper—with deterministic ordering and normalized-text deduplication. Preparation and acceptance deadlines default to 150 ms and provider-status reads to 300 ms; both fail open, so a stuck provider never blocks the agent. Recalled content is always labeled untrusted background information and is never elevated to authorization or higher-priority instructions. All configuration values (maxItems 1–6, maxBytes 1–4000, two timeout fields 1–5000 ms) are validated integers and can be tuned down further.

Best suited for users who already run Memory and MSE in their Harness profile and want both providers to recall cooperatively within a single message, with a clear Settings overview of each provider state. If you prefer a bounded, configurable, non-intrusive integration over yet another standalone tool, Brain quietly merges two memory streams into one source-attributed context—without deleting or migrating existing data, and provided no other plugin already exposes missherBrain.