📖 User Guide · 🧪 Training Roadmap · Changelog
--- ## What this is Most of the time, what blocks us isn't "we don't know how." It's "we can't say what we want." You drop a line like *"just make that part better, it feels off"* and the agent can only guess. Guessing right is luck. Guessing wrong costs a full round of rework. **The wasted tokens and your wasted time both go into guessing.** Memory Fitting flips the order: > **Instead of letting the agent guess, let it put up targets first — then ask.** It writes down 3-5 directions you might want, then asks in batches about **the places those directions still can't be told apart**. After each round it re-weighs every direction and **puts the updated list in front of you**. Once it's clear enough, it proposes: *"I read this as X — right?"* **You never have to describe what you want. You only have to correct its guess.** And correcting is far cheaper than describing. --- ## How it works ``` ① Predict The agent writes down 3-5 directions you might want (each with: what it is, and what would be true if it's right) ↓ ② Compose Design 2-4 questions for THIS round, aimed at the differences between directions — not at what's already known ↓ ③ Ask await ctx.userQuestions.ask({ questions: [...] }) One batch in, all answers back ↓ ④ Think Digest the round → update directions (promote / drop / add / merge) 【SHOW IT TO THE USER】← this step is not optional ↓ One direction clearly ahead and stable? │ │ no yes │ ↓ └──► back to ② ⑤ Propose: "I read this as X" (+ the runners-up) ↓ ⑥ You confirm → archive ``` ### Why step ④ cannot be skipped | | Without showing | With showing | |---|---|---| | The agent's claim | "I get it" — an unverifiable self-report | "Direction 3 leads; direction 1 is ruled out" — **verifiable** | | What you do | Wait, and find out only if it's wrong | See at a glance whether it drifted, correct on the spot | **Putting the agent's understanding in plain sight is the only substantive difference between this and "the agent asked a couple of things and then said it understood."** --- ## Interface A floating panel in the bottom-right: **Settings** (default) · **Archive** · **Start a fitting**. ### Positioning: its own column, so nothing overlaps The bottom-right of DSH is already a stack (dsh-cua and Ark9Canvas each probe for the other's FAB, but are **unaware of each other**): | Plugin | FAB | Panel | |---|---|---| | dsh-cua | right:16 bottom:96 | right:16 bottom:152 | | Ark9Canvas | right:16 bottom:96\|148 | right:16 bottom:152\|204 w-440 | | **This plugin** | right:16 **bottom:16** | **right:472** bottom:16 · height min(78vh,660px) | Our FAB sits at the very bottom of the stack; our panel moves to the **left of** the right-hand column and spans full height — completely clear of the \`right:16\` column, **no overlap whether or not the others are open**. On narrow screens (<900px) it degrades to a centered overlay. --- ## Switches: all off by default This plugin makes a point of not getting in your way. All three "contamination surface" switches are **off by default**: | Switch | Default | Meaning | |---|---|---| | \`injectContext\` | **off** | Inject one very short note each turn so the model knows it can start a fitting | | \`writeMemory\` | **off** | Write the converged conclusion into dsh-auto-memory | | \`exposeTools\` | **off** | Expose \`memory_fit_*\` tools to the model (registering costs tool-schema context every turn) | | \`localArchive\` | **on** | The plugin's own local archive | **Why off by default**: on a shared API, every turn of injection costs token rent — and the more often you inject, the more the model's attention to it decays. The right strategy isn't "inject more," it's "make every injection count." So: no injection unless you ask for it. You don't need to touch the switches to use it: click **Start a fitting** in the panel. No tools need to be visible to the model. --- ## Where the data lives ``` ~/.dsh/memory-fitting/ config.json configuration sessions/