DS4Server/PLAN.md

# DS4Server implementation plan

Bundle/application identifier: `DS4Server.rfc1437.de`

The GUI is a Rust/Iced reimplementation of the user-facing model server and
coding-agent workflows in `../ds4`. The existing engine remains the behavioral
reference: `ds4.c`/`ds4.h` define model and session behavior,
`ds4_server.c` defines the compatible API, and `ds4_agent.c` defines agent
sessions, tools, and turn orchestration.

## Current status

| Area | Status | Available now | Still open |
| --- | --- | --- | --- |
| App shell | Implemented | Native multi-window Iced app, Codex-inspired layout, native Application/Edit/Window menus, Preferences panel, Model Manager, and streaming composer | Native app release packaging |
| Projects and sessions | Implemented with transcripts | Native folder picker, project-name dialog, create/select/delete projects and sessions, and durable structured chat history | KV payloads, rename/archive, and model binding |
| Persistence | Implemented for metadata, preferences, and chat | SQLite in Application Support, Diesel ORM, embedded migrations, constrained defaults, streamed message updates, and reopen tests | Downloaded-artifact metadata and session-runtime migrations |
| Model runtime | DeepSeek Flash vertical implemented | Rust-owned mmap/model/session graph, reused Metal kernels, 2K compressed-attention generation, DS4 sampling, lazy background loading, cancellation, and idle unloading | Ratio-4 sparse indexer beyond 2K, DSpark/SSD/steering, GLM/Pro, KV checkpoint persistence, and shared acquisition for HTTP |
| Local endpoint | Not started | Nothing listening | OpenAI-compatible HTTP and streaming surfaces |
| Agent | Basic local chat implemented | Multi-turn role rendering, structured reasoning disclosure, streamed DeepSeek Flash text, composer send/stop, and durable transcript rehydration | Tools, approvals, compaction, statistics, and KV-backed resume |
| A2UI | Future extension | bDS2 provides the reference structured render-tool contract | Native inline surfaces for local chat after core chat and tools are stable |
| Dev brain | Not started | No vault access | Obsidian vault selection, durable access, and agent knowledge/memory tools |
| Release | Not started | Development binary builds and tests | `.app` packaging, signing, entitlements, and notarization |

The application manages durable project/session metadata and model preferences.
Its Model Manager downloads, validates, and deletes managed GGUF artifacts
without blocking the UI. Local chat now lazily acquires the Rust DeepSeek Flash
executor, streams generated text, supports cancellation and multi-turn role
replay, persists reasoning and answers while they stream, rehydrates selected
sessions, and unloads model resources after the configured idle timeout. The
current Rust graph supports DS4 compressed attention through 2K context; the sparse ratio-4 indexer,
KV checkpoints, HTTP service, and agent tools remain open.

## Phase 0 — project and session shell

Status: **implemented**.

- Show projects in a sidebar with their sessions nested underneath.
- Create projects with a native macOS folder picker and a project-name dialog.
- Create, select, and delete projects and sessions.
- Use a Codex-inspired two-pane layout with embedded, action-specific SVG icons.
- Persist projects and sessions in SQLite under
  `~/Library/Application Support/DS4Server.rfc1437.de/`, using Diesel ORM
  queries and embedded Diesel migrations.
- Never delete workspace contents when removing app metadata.

Exit criterion: restart the app and see the same project/session tree.

## Phase 1 — model library and on-demand loading

Status: **partially implemented**. The typed preference/download/intake path and
a DeepSeek Flash Rust/Metal vertical are implemented, including compressed KV
through 2K context, DS4 sampling, a single background owner, cancellation, and
idle unload. Sparse indexed attention, persistent KV/session sync, other model
families, and optional execution modes remain open.

1. **Implemented:** the typed model/runtime preference contract described below
   is complete before chat, the HTTP endpoint, or the engine lifecycle.
2. **Partially implemented:** mmap-backed loading, tokenizer/prompt rendering,
   Rust-owned Flash session creation/evaluation/sampling, and model ownership
   are complete. Prefix sync plus KV save/load remain.
3. **Implemented:** retain the tested model restrictions from DwarfStar. Main
   and DSpark artifacts reject invalid GGUF versions, catalog mismatches, and
   incompatible metadata, tensor shapes, offsets, or quantization before
   promotion and again before model use.
4. **Partially implemented:** reuse the existing `.metal` kernels and preserve mmap-backed resident
   loading plus the explicit SSD-streaming path. Keep Objective-C only at the
   Metal interop boundary.
5. **Implemented for local chat:** keep one engine resident at most, matching DwarfStar's instance-lock and
   memory assumptions.
6. **Implemented for local chat:** move engine work off the UI thread and support cooperative cancellation at
   the safe session boundaries already defined by `ds4_session_sync`.

### DS4 KV-cache port and shared session core

KV-cache behavior is part of the DS4 engine contract, not a new application
cache to redesign. Port the existing `ds4_session` and `ds4_kvstore` behavior
to Rust while preserving its semantics and on-disk compatibility:

- A live engine session owns one mutable inference timeline: exact rendered
  tokens, logits, and the DS4-specific KV graph state. Callers provide the full
  token prefix and the Rust port of `ds4_session_sync` must retain the longest
  compatible prefix, evaluate only its suffix, or safely rewind, invalidate,
  and rebuild exactly when DS4 does.
- Carry over DS4's safe cancellation/checkpoint boundaries, common-prefix
  matching, canonical rewrite and replay, snapshots, payload staging,
  save/load, strip-to-transcript behavior, and atomic replacement. The engine
  continues to own the opaque KV payload format; UI, HTTP, and persistence code
  must not reinterpret tensor state.
- Port `ds4_kvstore` policy rather than replacing it with a generic cache:
  exact token and rendered-text prefix lookup, cold and continued aligned
  checkpoints, model/quantization/context/payload-ABI compatibility checks,
  tool-call replay metadata, disk-budget enforcement, hit tracking, and
  eviction behavior.
- Use one process-wide Rust session registry and KV checkpoint store for both
  local chat and the HTTP server. A remote request bound to an application
  session resolves to the same logical DS4 session as the GUI and mutations are
  serialized so two callers never write one live timeline concurrently.
  Stateless HTTP requests still participate in the same compatible-prefix
  checkpoint pool, so a prefix produced locally can be reused remotely and
  vice versa.
- Store that shared checkpoint pool under Application Support in
  `kv-cache/`, using the DS4 key/header/payload rules. Session rows hold only an
  optional reference to their current checkpoint; the large opaque payload is
  not copied into separate GUI and HTTP caches.
- Sharing does not merge unrelated conversations or expose one session's
  transcript to another. It shares the DS4 implementation, checkpoint pool, and
  explicitly bound logical session; each unrelated live conversation keeps its
  own mutable session state.
- Engine unload must first leave every reusable live timeline in a valid
  persisted checkpoint. Reload restores a compatible checkpoint immediately;
  if model identity, quantization, context size, payload ABI, or rendered
  history is incompatible, rebuild from the transcript and replace the stale
  checkpoint only after the new state is valid.

### Lazy lifecycle

- Start the application and the local HTTP endpoint with no model loaded.
- Acquire the configured model only when generation starts, whether initiated
  by a local chat or an OpenAI-compatible endpoint request. Concurrent first
  requests share the same load instead of opening the model more than once.
- Expose `Unloaded → Downloading → Loading → Ready → Unloading` plus failure
  states and progress in the GUI. A request that triggered loading waits for
  the shared load and receives a clear error if download or initialization
  fails.
- Treat active local generations, endpoint requests, prefills, and tool turns
  as model activity. Start the idle timer only when all activity has stopped.
- Unload the engine after the configured idle timeout. New work cancels a
  pending unload; work arriving after unload transparently loads the model
  again. Persisted transcripts and KV files remain available when model memory
  is released.
- Changing the selected model prevents new work from using the old model and
  unloads it as soon as current work reaches a safe boundary.

### Preferences foundation (implemented)

- Provide a Preferences panel from the sidebar and the standard
  macOS `Command-,` shortcut.
- Let the user choose the active model. Default to `deepseek-v4-flash`.
- Let the user enable or disable DSpark for the selected model. Disable the
  control for models that do not support DSpark.
- Let the user configure the inactivity timeout; default to 10 minutes.
- Persist preferences in the application SQLite database. Changing preferences
  never loads a model by itself.

### DS4 model and generation preferences (implemented)

The Preferences panel must expose every persistent model, runtime, and
generation parameter accepted by the `ds4` command line. `../ds4/ds4_cli.c`,
`ds4.h`, and the shared GPU/distributed parsers are the source of truth; use the
same defaults, ranges, units, dependencies, and model-family overrides rather
than inventing application-specific behavior.

- **Model and capacity:** selected main GGUF (`--model`), optional MTP/DSpark
  support GGUF (`--mtp`), context tokens (`--ctx`), maximum generated tokens
  (`--tokens`), and system prompt (`--system`). Managed catalog selections map
  to their installed files instead of requiring users to type paths.
- **Sampling and reasoning:** temperature (`--temp`), top-p, min-p, optional
  seed, and one reasoning mode representing `--think`, `--think-max`, or
  `--nothink`. Preserve DS4's explicit-vs-default distinction because GLM
  applies model-family sampling defaults only when the user has not overridden
  them.
- **Execution:** the macOS app uses Metal exclusively; omit the unnecessary CPU
  backend and backend selector. Persist CPU helper threads for host-side work,
  GPU power percentage, prefill chunk, exact/quality kernels, and warm weights.
- **Speculative decoding:** MTP draft-token count and margin, GLM MTP, GLM MTP
  timing, DSpark enablement, DSpark confidence threshold, and DSpark strict
  target-only decode. Enabling a dependent control must enable or require its
  support artifact exactly as the CLI does.
- **SSD streaming:** enablement, cold start, expert count or GiB cache budget,
  fully resident GLM layers, and explicit expert preload count. Retain the
  distinction between unset/automatic and an explicit zero where DS4 does.
- **GPU placement:** per-device VRAM budgets, device indices, and CUDA tensor
  parallelism when available. Hide or disable these controls in a Metal-only or
  CPU-only build rather than accepting values the engine cannot honor.
- **Directional steering:** direction-vector file, FFN scale, and attention
  scale, including the CLI rule that a file with no explicit scale defaults the
  FFN scale to 1.
- **Advanced diagnostics that change model execution:** simulated used memory
  and routed expert profile output. Keep these in an Advanced section with
  clear diagnostic labels, but persist and pass them through like their CLI
  equivalents.

Prompt sources, raw one-shot mode, inspect/dump/perplexity/imatrix/self-test
actions, and distributed coordinator/worker network topology are CLI workflows,
not persistent model parameters; they belong to their corresponding chat,
diagnostic, or deployment surfaces rather than Preferences. Any future DS4 CLI
option must be classified explicitly as either a preference or one of these
non-preference actions.

Implement this before chat as one typed settings path shared by the local GUI,
the lazy engine lifecycle, and HTTP defaults:

- Persist unset/automatic values separately from explicit values and provide a
  Reset to DS4 defaults action. Validate before saving with the same accepted
  ranges and cross-field rules as the CLI.
- Mark settings as load-time or turn-time. A load-time change unloads/reloads at
  the next safe boundary; sampling, seed, token budget, and reasoning changes
  apply to the next turn without duplicating the engine.
- Local chat uses these values directly. HTTP requests use them as defaults and
  may override only request-scoped generation fields supported by the API; both
  paths still resolve through the same effective-settings builder.
- Add a focused parity test that inventories every option branch in the DS4 CLI
  model/runtime, sampling, and steering groups and fails until each option is
  mapped to a typed preference or explicitly classified as a non-preference
  action.

Exit criterion: save, restart, and recover every supported DS4 model/runtime
and generation value; show the same effective defaults and validation as the
CLI; and construct one engine configuration plus one turn configuration without
chat or HTTP code re-parsing preference fields.

### Targeted downloads and storage

- Preferences only selects the model/runtime configuration. A separate Model
  Manager window, reachable from the native Window menu, lists every managed
  main and DSpark artifact with its on-disk size and state, and provides
  Download/Resume, Validate, Stop, and confirmed Delete actions.
- A dedicated worker thread performs HTTP and verification work so the Model
  Manager and the rest of the application stay interactive.
- Use an in-process Rust HTTP client with Rustls rather than invoking a system
  downloader. Stream response chunks directly into a `.part` file and use a
  Rust SHA-256 implementation for verification.
- Report aggregate bytes present, total bytes, bytes missing, percentage,
  smoothed transfer rate, current artifact/verification phase, and estimated
  time remaining. Keep the detailed display in Model Manager and a persistent
  status bar with a Stop action while the transfer runs.
- During SHA-256 verification, show a dedicated progress bar based on bytes
  read and hashed rather than the already-complete file size. Update the bytes
  verified, verification throughput, percentage, and estimated time remaining
  throughout the pass so large models never appear stalled at 100%.
- Stop is non-destructive: close the active transfer and retain its `.part`
  file. Starting the download again in the same process or after quitting and
  relaunching must derive progress from that file and issue an HTTP range
  request from its exact byte length. If the server ignores the range request,
  safely restart that artifact from byte zero rather than appending corrupt
  data.
- Promote a `.part` file only after its exact size and SHA-256 are verified. A
  stopped verification keeps the complete partial file so the next run can
  verify and promote it without downloading it again.
- Store managed model artifacts under
  `~/Library/Application Support/DS4Server.rfc1437.de/models/<model-id>/` and
  mmap/load them from there. Do not place managed models in the app bundle or a
  project directory.
- Download only the artifact explicitly requested in Model Manager; never
  prefetch other catalog models or optional artifacts.
- Remove checksum-invalid partial files without touching a valid model. Keep
  partial files after cancellation, application exit, network failure, or an
  interrupted verification so they remain restart-resumable.
- Delete model artifacts only through an explicit, confirmed Model Manager
  action; changing Preferences must never remove files.

Exit criterion: select a supported DeepSeek V4 or GLM 5.2 GGUF, load it without
blocking the UI only when a chat begins, run a deterministic prompt, unload it
after the configured idle timeout, and pass the matching `../ds4` token-output
fixture. Repeat through the local endpoint and verify both paths share the same
engine lifecycle.

## Phase 2 — OpenAI-compatible local endpoint

Status: **open; not started**.

1. Add a localhost-only HTTP service whose lifecycle is independent of the
   engine. It can listen while the model is unloaded and acquires the configured
   model only for inference. Start and stop it from the GUI; never expose a LAN
   listener without an explicit setting.
2. Implement the DwarfStar compatibility surface in this order:
   - `GET /v1/models`
   - `POST /v1/chat/completions`
   - `POST /v1/responses`
   - `POST /v1/completions`
   `GET /v1/models` reports the configured model without forcing it to load.
3. Support streaming SSE, reasoning output, usage accounting, cancellation,
   sampling parameters, tool schemas, and tool choice.
4. Port exact sampled tool-call replay and deterministic canonicalization from
   `ds4_server.c` so a client's normalized JSON does not destroy KV-prefix
   reuse.
5. Route API conversations through the shared session registry and KV store
   defined in Phase 1. Responses `conversation` bindings and any other
   persistent request identity map to the same application session used by the
   GUI; stateless requests use shared DS4 prefix lookup. Persist/restore only
   through the engine-owned serialization format. Add resident batching only
   after single-session and cross-surface correctness are established.
6. Display endpoint address, model, active requests, token rates, and errors in
   the GUI.

Exit criterion: the upstream DwarfStar server tests pass against the app for
non-streaming and streaming Chat Completions and Responses calls, including a
multi-turn tool call. A compatible prefix created by local chat must register a
remote cache hit, a remote checkpoint must be reusable by local chat, and a
GUI/HTTP race on one bound session must serialize without corrupting its token
or KV frontier.

## Phase 3 — durable agent sessions

Status: **partially implemented**. Project/session metadata and structured
transcripts are durable; model binding and KV payload persistence remain open.

Extend each metadata-only session with transcript and shared-checkpoint
metadata:

```text
Application Support/DS4Server.rfc1437.de/
  data.sqlite3
  models/<model-id>/
    main.gguf
    dspark.gguf       # only when enabled
  kv-cache/
    <ds4-cache-key>.kv
```

- **Partially implemented:** migrated message tables persist user text,
  reasoning state, and assistant output while generation streams. Add the
  optional current-checkpoint key to SQLite. Keep large engine-owned KV
  payloads only in the process-wide Phase 1 store, written with atomic
  replacement; do not introduce separate agent-only or HTTP-only cache formats.
- Record model identity, context size, rendered token history, title,
  timestamps, and working directory.
- Restore compatible KV immediately. If KV is absent or incompatible, rebuild
  from the transcript and show prefill progress.
- Add strip/archive behavior equivalent to the native agent: retain the
  transcript and clear its checkpoint reference. Remove the cache file only
  when it is not reusable by another live/session prefix; otherwise normal DS4
  disk-budget eviction reclaims it.
- Version formats before the first release and add migrations only when a real
  format change exists.

Exit criterion: quit during normal use, relaunch, select a session, and resume
the same conversation without prefill when its KV payload is compatible.

## Phase 4 — agent chat and tools

Status: **basic local chat implemented**. The conversation area streams
multi-turn DeepSeek Flash output, persists and rehydrates structured reasoning
and answers, and supports Stop; tool turns and the coding-agent runtime remain
open.

Every chat feature must persist its semantic state and rehydrate it when a
session is reopened in the same implementation slice. In-memory-only chat
features are incomplete by definition; disclosure state and other purely
ephemeral presentation preferences are the only exception.

1. **Partially implemented:** build the classic chat surface. Transcript,
   structured reasoning disclosure, composer, streamed output, and Stop are
   complete; tool-call cards, queued input, prefill progress, and generation
   statistics remain. Local turns must call the same session registry and KV
   path used by the HTTP endpoint.
2. Port the native agent turn loop from `ds4_agent.c`, including model-specific
   system prompts and DeepSeek DSML/GLM tool-call parsing.
3. Implement the local tool set with the project directory as its boundary:
   - shell command execution
   - file read and continuation
   - file write
   - anchored edit
   - text/file search
   - opt-in Dev Brain knowledge lookup and memory storage
4. Stream assistant tokens and partial tool calls into Iced while inference and
   tools run on workers. Preserve cooperative interruption and a valid KV
   prefix at every stop point.
5. Require confirmation for commands or writes outside the project boundary,
   destructive actions, and network tools. Add web search/page visiting only
   after the local tool loop is stable.
6. Port context compaction and system-prompt reinjection after ordinary
   multi-turn operation is correct.

### Native macOS menus (required with chat)

The application must install and maintain a complete native macOS menu bar;
the application-name menu alone is not sufficient. Use platform menu roles,
ordering, selectors, and standard key equivalents rather than drawing menus
inside the Iced window.

- Provide the standard **Application**, **File**, **Edit**, **View**,
  **Window**, and **Help** menus. Include the normal macOS Application actions
  (About, Settings, Services, Hide, Hide Others, Show All, and Quit) and Window
  actions (Minimize, Zoom, Bring All to Front) with native behavior.
- Make File actions operate on real application concepts: New Session, Add/Open
  Project, and Close Window. Make View actions cover actual available views,
  including sidebar visibility and Enter Full Screen. Help opens the local help
  or documentation surface. Do not add permanently inert placeholder items.
- The Edit menu is part of the chat acceptance criteria. Undo, Redo, Cut, Copy,
  Paste, Paste and Match Style where supported, Delete, and Select All must act
  on the current first responder. They must work in the chat composer and every
  other editable text field; Copy and Select All must also work for selectable
  transcript text.
- Prefer native first-responder selectors for native controls. Where an Iced
  widget owns focus, bridge the same menu command into that focused widget so
  menu clicks and `Command-Z`, `Shift-Command-Z`, `Command-X`, `Command-C`,
  `Command-V`, and `Command-A` behave identically.
- Validate menu enabled state whenever focus or application state changes:
  editing commands reflect the focused selection/clipboard and session/project
  commands reflect whether their action is currently possible. Menu actions
  must use the same application messages as toolbar/sidebar actions instead of
  duplicating business logic.
- Preserve conventional macOS labels, grouping, keyboard shortcuts, and
  accessibility metadata. Keep platform-specific menu integration behind the
  macOS application shell so the chat and session core remain platform-neutral.

Exit criterion: with the composer focused, every standard Edit command works
from both the menu and its keyboard shortcut, including undo/redo across
multiple edits; transcript text can be copied; and Application, File, View,
Window, and Help expose functional, correctly enabled native actions.

### Dev Brain — Obsidian knowledge and memory

- Add an optional Dev Brain setting in Preferences. Use the native macOS folder
  picker to select an existing Obsidian vault, allow the vault to be changed or
  disconnected, and persist its canonical path plus a security-scoped bookmark
  in SQLite so sandboxed builds can regain access after restart.
- Keep the vault global to the application rather than tied to one project or
  session. The selected vault remains in its original location; never copy it
  into Application Support or a project directory.
- Add one local `dev_brain` agent tool with the minimum useful operations:
  search Markdown notes, read a selected note, and create or append a memory
  note. Return vault-relative paths with results so the agent can cite where
  knowledge came from.
- Restrict every operation to the explicitly selected vault. Resolve and check
  paths before access, ignore `.obsidian` and hidden files, and do not provide
  deletion or arbitrary overwrite operations initially.
- Let the agent retrieve relevant global knowledge during a turn and record
  durable conclusions after a turn. Vault access is opt-in, does not preload the
  model, and must not block sessions when no vault is configured or available.
- Reuse the existing local file/search implementation underneath the new vault
  boundary instead of building a second filesystem stack.

Exit criterion: create a project session, ask the model to inspect and edit a
small fixture repository, review each tool result in the GUI, restart, and
continue from the persisted session. With Dev Brain enabled, store a memory in
one session and retrieve it with a cited vault-relative note path from a session
in another project after restarting the app.

## Phase 5 — A2UI rich local-chat surfaces

Status: **future extension; not part of the core milestone**. Start only after
ordinary local chat, structured tool calls, transcript persistence, and
cancellation are stable. Use the A2UI implementation in `../bDS2` as the
behavioral reference, especially its render-tool schemas, typed surface
normalization, inline message placement, chart geometry, and interaction state.

- Expose typed `render_card`, `render_chart`, `render_form`, `render_list`,
  `render_metric`, `render_mindmap`, `render_table`, and `render_tabs` tools to
  the local agent. Keep these tools presentation-only; they return structured
  data and do not perform project mutations themselves.
- Support the bDS2 chart set initially: bar, stacked bar, line, area, pie,
  donut, and heatmap. Render charts and mind maps with native Iced primitives
  and SVG geometry inside the assistant message rather than adding a web view.
- Treat tool arguments as untrusted data. Validate type, nesting depth, row and
  series counts, numeric values, labels, action names, and payload size before
  rendering. Never accept model-authored HTML, JavaScript, styles, file URLs, or
  arbitrary commands.
- Attach each validated surface to the assistant message/tool call that created
  it and persist the versioned structured payload with the transcript. A
  restarted session must reconstruct the same surface without asking the model
  to emit it again.
- Build a surface only after its tool arguments are complete and valid while
  allowing ordinary assistant text to continue streaming. Malformed or unknown
  render payloads fall back to a bounded JSON/text representation and must not
  fail the surrounding chat turn.
- Preserve local interaction state for forms and tabs. Dispatch form submits
  and card actions through a small allowlist of application actions; any action
  that mutates files, settings, or external state still follows the normal
  approval path.
- Provide keyboard navigation, readable labels, theme-aware colors, and a text
  or tabular fallback for chart data so the information remains accessible and
  copyable.
- Keep A2UI local-chat-only initially. Do not extend the OpenAI-compatible HTTP
  surface or invent a general plugin/component protocol until a real client
  requires it.

Exit criterion: ask the local agent to present fixture data as each supported
surface type; verify a chart and nested tab surface render correctly, an
interactive form returns validated values, malformed payloads degrade safely,
and all surfaces survive application restart with their message history.

## Verification and packaging

Status: **partially implemented**. Formatting, Clippy, build, and unit-test
commit gates are documented in `AGENTS.md`; database migration/CRUD constraints
and the preferences shortcut have unit coverage. Model fixtures, integration
tests, and release packaging remain open.

- Keep unit coverage narrow: persistence round trips, format compatibility,
  DS4 KV prefix/rewrite/store/load parity, cross-surface session binding,
  request parsing, and tool-boundary checks.
- Reuse `../ds4` fixtures for prompt rendering, sampling, server streaming, and
  KV correctness; run live Metal tests only when a supported GGUF is available.
- Preserve the DwarfStar/llama.cpp notices required by `../ds4/LICENSE` for
  reused or adapted source and kernels.
- Produce a signed `.app` with the exact identifier above, then add hardened
  runtime, entitlements, notarization, and update delivery as release work.

## Recommended next milestones

1. Finish the Flash session core as one larger slice: port the ratio-4 indexer
   and indexed attention for full configured context, then add prefix sync and
   compatible KV checkpoint save/load.
2. Put the OpenAI-compatible endpoint on the same engine lifecycle so local
   chats and HTTP requests cannot load duplicate engines.
3. Add transcript/KV persistence, then connect the existing conversation UI and
   finally port the agent tools.
4. Add the opt-in Dev Brain tool after local file/search boundaries and agent
   approvals are stable.
5. Add bDS2-style A2UI render tools and native inline surfaces after the core
   local chat/tool loop is complete.