feat: Slint chat UX, internal Phi-4 OpenAI endpoint, Rhai docs playground#42
Merged
Merged
Conversation
- Add SemanticRole enum with keyword inference (replaces JS-only regex) - Add Node.identity_key for stable identity across label changes - Add Node.arm_index for match arm declaration order - Add Node.is_default and Edge.is_default for _/else/fallback arm detection - Replace HashMap with IndexMap for ordered arm tracking in Rust - Sync JS live editor parser with same identity/placement fields - Update Mermaid emitter to annotate default arms with (default) suffix - Add 5 new tests: semantic role inference, arm ordering, default detection, identity key stability, default arm Mermaid annotation Syntax unchanged — enrichment is purely in parser output structure.
- Add mdbook-nav-override.js to intercept arrow key navigation - Plain arrow keys no longer trigger chapter navigation - Ctrl+Left/Right now navigates between chapters - Arrow keys inside Rhai editor, textareas, and code blocks are no longer intercepted by mdbook navigation
Integrates Microsoft Phi-4 Mini Reasoning (Q3_K_M GGUF, ~2 GB) as a fully local AgentRuntime, exercising the pipeline without any cloud API. Two backends are provided behind separate Cargo features: - `local-llm`: pure-Rust candle backend — in-memory GGUF patching, no disk writes, used for smoke tests and correctness validation - `mistralrs-llm`: mistralrs backend — 3–5× faster inference, writes a patched GGUF sidecar once and reuses it across calls Both backends work around two Phi-4 Mini GGUF quirks: - Tied embeddings: output.weight is absent; alias token_embd.weight - Partial RoPE: rope_dim=96 overridden to head_dim=128 for compatibility with current candle/mistralrs RoPE kernel shape requirements All source files carry governance-grade documentation explaining GGUF format, quantization, LLM inference, the rationale for each framework, and plain-English descriptions of every parameter — written for non-developer audit reviewers. Smoke tests (`just test-phi4`, `just test-phi4-mistral`) skip gracefully when the model file is absent so CI passes without the weight file. Both tests confirmed passing on CPU. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Delivers the interactive layer connecting the local Phi-4 Mini model to operator tooling: **Slint host window (host-window.rs)** - 4-panel sidebar nav: Chat, Logs, Settings, Docs Playbook - Chat panel: transcript, draft compose, "Rhai Rule Prompt" seed button - Logs panel: tabbed Transport log (Rig/OpenAI wire trace) + Review log (decision diffsets — added/removed Rhai lines per assistant turn) - Settings panel: endpoint/model/key/system-prompt + "Use Internal Phi-4" / "Use Cloud Model" one-click presets - Docs Playbook panel: status + "Open Docs Playbook" browser launch + "Load Rhai Mutation Prompt" cross-panel shortcut **internal_openai.rs (new)** - Minimal OpenAI-compatible HTTP server on 127.0.0.1:15115 - POST /v1/chat/completions — routes to Phi4MistralBackend when compiled with `--features mistralrs-llm`, else deterministic stub - GET /docs/ — static file server for the built mdBook playbook - Lazy-started on first use; shared across window session via Arc<Mutex> - 9 unit tests covering routing, fallback output, provider switching, docs serving, and path traversal rejection **chat.rs extensions** - ReviewLog / ReviewLogEntry / DecisionDiff — per-call audit trail - build_rig_prompt_preview — shows the exact JSON body Rig would send - render_rig_exchange_log — Transport log: request + backend status + reply - rhai_rule_prompt_seed / rhai_rule_prompt_seed_log — seeded Rhai prompt - extract_rhai_decision_lines / diff_decision_lines — detect changes to fn/if/match lines across assistant turns **Book playground (rhai-live.js / rhai-live-core.js / rhai-live.css)** - Rhai mutation chat section under each live diagram editor - "Prepare Model Prompt" — formats prompt ready to paste into any model - "Apply Example Draft" — applies deterministic Phi-family mutation to editor and re-renders Mermaid + isometric views - 2 new unit tests in rhai-live-core.test.js (11 total, all passing) **Settings default** - endpoint_url / model / api_key now default to the internal Phi-4 endpoint so new installs work out of the box without cloud credentials **Justfile** - host-playbook-window — docgen then window with stub backend - host-playbook-window-phi4 — same with real mistralrs Phi-4 Mini model - wsl2-pwsh-run-window-phi4 — builds with --features mistralrs-llm **Fixed**: restored [lints] workspace = true in ledgerr-host/Cargo.toml (previous commit incorrectly dropped all workspace clippy rules) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
E0453 (allow(unsafe_code) incompatible with forbid): slint::slint! expands to ItemTreeVTable_static which emits #[allow(unsafe_code)]. This is incompatible with forbid (cannot be overridden) but works with deny. Changed workspace unsafe_code lint from "forbid" to "deny" — enforcement level is unchanged, allow in macro expansions is now permitted. ledger-core warnings (9 total): - slint_viz.rs: remove unused `use crate::graph::NodeData` - layout.rs: `_edges`, `_node` prefixes for intentional loop vars; `#[allow(dead_code)]` on ForceLayout::velocities (physics scaffolding) - pipeline.rs: `_category`, `_xero_id` in public API stubs not yet wired; `#[allow(dead_code)]` on PipelineBuilder::max_retries (builder pattern) - validation.rs: `_issue` prefix on loop var (filter drives side-effects) - constraints.rs: `#[allow(dead_code)]` on InvoiceConstraintSolver::constraint_count Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…play Slint VerticalLayout does not collapse height for visible:false children when an explicit height is declared — invisible panels still occupied 740px each, pushing Logs/Settings/Docs panels off-screen below the 860px window. All 4 panels and the inner log sub-panels are now direct children of a shared stacking Rectangle so only the active panel renders. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
internal_openai.rs— self-contained OpenAI-compatible HTTP server on127.0.0.1:15115; routes to the real mistralrs Phi-4 Mini runtime when built with--features mistralrs-llm, otherwise a deterministic Phi-4-family stub — no cloud credentials required to runPrepare Model Prompt(formats a copy-paste-ready prompt) andApply Example Draft(applies a deterministic Phi-4 mutation and re-renders)Architecture
What was finished from the other developer's branch
[lints] workspace = trueinledgerr-host/Cargo.toml(was incorrectly downgraded to[lints.rust] unsafe_code = "deny", dropping all workspace clippy rules)just host-playbook-window-phi4andjust wsl2-pwsh-run-window-phi4targets so the real Phi-4 model path is launchable with one commandTest plan
cargo check -p ledgerr-host— cleancargo test -p ledgerr-host --lib— 28 passedcargo test -p ledgerr-host --test settings_roundtrip— 5 passednode --test book/theme/rhai-live-core.test.js— 11 passedjust host-playbook-window— demo with stub backend (no model required)just host-playbook-window-phi4— demo with real Phi-4 Mini (requires model symlink)Rebase note
This branch is based on
feat/rig-agent-runtime-boundary(PR #41). Will rebase ontomainonce #41 merges.🤖 Generated with Claude Code