AI Agent Hub
Back to plugins
🖥️

dsh-cipher

Client Updated 2026.08.29

Run the following command in DeepSeek Harness:

dsh plugin install NoxTyrannus/dsh-cipher

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install NoxTyrannus/dsh-cipher in the DeepSeek Harness terminal; the full source is available at https://github.com/NoxTyrannus/dsh-cipher

About this plugin

The Standard, PTC, and Minimal presets in DSH are all stateless: the moment a conversation ends, every accumulated insight, preference, and piece of context is gone. dsh-cipher ports the core of cipher's design—four local memory types (attention, experience, preference, cognitive) and the UNNI interactive / LOOP memory-driven session modes—into a native DSH plugin bundle. No Rust process, no cipher binary. After install, UNNI and LOOP show up immediately in the native agent-preset picker, right alongside Standard.

After every turn, four memory agents fire in parallel (one LLM call each, error-isolated) to sediment evidence into the four memory types with deduplication and per-session watermark control. On the next turn or in a later session, deterministic recall—per-type quotas, 14-day time decay, and term/keyword matching—injects relevant memories into the system prompt, re-recording only when the snapshot actually changes so provider KV caches stay reusable. LOOP sessions re-arm themselves with the freshest attention and experience memory after each sediment round, guarded by a minimum interval and a per-session run cap, then wrap up quietly once converged or stopped by the user. The plugin also ships an editable SOUL, a collapsible AgentPool sidebar showing live subagent status, a per-session Cipher mode toggle in the session header, and two model tools: cipher_memory_search and cipher_memory_ingest.

If you already work in DSH and want conversations that remember your domain preferences, past decisions, and working habits across sessions—or an agent that keeps iterating along its own memory trail without you manually re-prompting every round—dsh-cipher packages that capability into a ready-to-pick preset, no separate infrastructure project required.

Use Cases

  • Retain user preferences, domain expertise, and past decisions across conversations instead of re-prompting from scratch each time
  • Let the agent iterate autonomously along its memory trail in LOOP mode, wrapping up quietly once converged
  • Accumulate four types of local memory—attention, experience, preference, cognitive—for long-running projects

Best For

  • DSH users who need persistent memory across sessions
  • Developers who want to reduce repetitive prompting and let the agent self-continue
  • Engineers building long-running conversational systems with provider KV-cache efficiency in mind