@polderlabs/bizar 3.23.1 → 4.2.0

package/README.md

@@ -7,7 +7,7 @@
 12 agents across 4 cost tiers. Odin routes, subagents execute, Forseti audits.
 
 [![npm](https://img.shields.io/npm/v/bizar?color=cb3837)](https://www.npmjs.com/package/bizar)
-[![RTK](https://img.shields.io/badge/rtk-integrated-8A2BE2)](https://github.com/rtk-ai/rtk)
+[![Headroom](https://img.shields.io/badge/headroom-integrated-8A2BE2)](https://github.com/headroomlabs-ai/headroom)
 [![Semble](https://img.shields.io/badge/semble-integrated-0ea5e9)](https://github.com/semble-ai/semble)
 [![Skills](https://img.shields.io/badge/skills.sh-integrated-f59e0b)](https://www.skills.sh)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
@@ -28,6 +28,7 @@
 - [Installation](#-installation)
 - [Provider Setup](#-provider-setup)
 - [Skill Discovery](#-skill-discovery)
+- [Memory Service](#-memory-service)
 - [Routing](#-routing)
 - [Self-Improvement](#-self-improvement)
 - [Contributing](#-contributing)
@@ -111,7 +112,7 @@
 - **Always parallel** — every request splits into 2+ simultaneous `task` calls
 - **Implementation splits across Thor + Tyr** frontend/backend, file split, impl+tests
 - **Forseti gates all Tier 4/5 work** — Tyr and Vidarr plans audited before execution
-- **Hindsight memory** — per-project banks with `bank_id: "<project-name>"`; default bank for general knowledge only
+- **Bizar Memory Service** — local Obsidian-compatible Markdown + Git-shared sync. Three namespaces: `projects/<id>/`, `global/bizar/`, `users/<id>/`. No external memory service required.
 
 ---
 
@@ -143,6 +144,8 @@ npx bizar
 
 The interactive installer walks you through component selection, agent choice, install mode, API key setup, and auto-restarts opencode.
 
+> **v4.0.0:** `@polderlabs/bizar` is now a single package — the dashboard server, the opencode plugin, and the typed SDK all ship inside it. No more separate `@polderlabs/bizar-dash` install.
+
 > **Windows users:** the `npm install -g @polderlabs/bizar` command above is the recommended path on Windows. The installer uses `irm | iex` for uv, `py -m pip` for the pip fallback, `taskkill` for forced kills, and JS `setTimeout` instead of `sleep` — so it works on both Windows PowerShell and POSIX shells. See the [Windows](#-windows) section below for prerequisites, known limitations, and the optional graph feature install.
 
 ### Source — git clone (Linux/macOS contributors)
@@ -198,41 +201,37 @@ bizar graph build
 - The `blessed` TUI library may render with quirks inside Windows Terminal; if you see garbled output in the TUI, use the browser UI via `bizar dash start --bg` and open http://localhost:4321 instead.
 - Some shell scripts under `config/skills/embedded-esp-idf/` are bash-only and require Git Bash (or WSL).
 - The bash `install.sh` does not run on Windows cmd/PowerShell; use the `npm install -g` path above.
-- RTK (Rust Token Killer) ships a bash-only installer. On Windows, install it manually via [Cargo](https://github.com/rtk-ai/rtk#installation) or a [prebuilt binary](https://github.com/rtk-ai/rtk/releases), then run `rtk init -g --opencode`.
+- Headroom ships a Python/npm installer. On Windows, install it manually: `pip install "headroom-ai[all]"` or `npm install -g headroom-ai`, then run `headroom wrap opencode`.
 
 ### Prerequisites
 
 - [opencode CLI](https://opencode.ai) installed and on `$PATH`
-- A [Hindsight](https://memory-api.polderlabs.io) API key for persistent memory
 - Provider connections (via `/connect` in opencode TUI)
-- [RTK](https://github.com/rtk-ai/rtk) (recommended) — CLI proxy that reduces LLM token consumption by 60-90%
+- [Headroom](https://github.com/headroomlabs-ai/headroom) (recommended) — CLI proxy that reduces LLM token consumption by 60-90%
 - [Semble](https://github.com/semble-ai/semble) (recommended) — AI-powered code search (used by Mimir agent)
 - [Skills CLI](https://www.skills.sh) (recommended) — Agent skill package manager (`npx skills add <owner/repo>`)
 
-### RTK Setup
+### Headroom Setup
 
-[RTK](https://github.com/rtk-ai/rtk) (Rust Token Killer) filters and compresses command output before it reaches the LLM context. It saves ~80% on common operations like `ls`, `git status`, `cargo test`, and `git diff`.
+[Headroom](https://github.com/headroomlabs-ai/headroom) filters and compresses command output before it reaches the LLM context. It saves ~80% on common operations like `ls`, `git status`, `cargo test`, and `git diff`.
 
 Install:
 
 ```bash
-# Homebrew (macOS)
-brew install rtk
-
-# Linux/macOS quick install
-curl -fsSL https://raw.githubusercontent.com/rtk-ai/rtk/refs/heads/master/install.sh | sh
+# Python (recommended)
+pip install "headroom-ai[all]"
 
-# Cargo
-cargo install --git https://github.com/rtk-ai/rtk
+# npm
+npm install -g headroom-ai
 ```
 
 Enable for opencode:
 
 ```bash
-rtk init -g --opencode
+headroom wrap opencode
 ```
 
-After setup, all Bash commands are transparently rewritten to their rtk equivalents — `git status` → `rtk git status`, `cargo test` → `rtk cargo test`. The LLM receives compact output, saving 60-90% on token costs.
+After setup, `headroom wrap opencode` injects config into `opencode.json` and starts a proxy. The LLM receives compact output, saving 60-90% on token costs. The wrapping is durable — all subsequent opencode sessions route through the headroom proxy.
 
 ### Semble Setup
 
@@ -294,6 +293,74 @@ This happens automatically and on-demand — agents self-discover capabilities w
 
 
 
+---
+
+## 🧠 Memory Service
+
+The **Bizar Memory Service** is a local-first, file-based memory subsystem that replaces the previously-disabled Hindsight MCP. Notes are Obsidian-compatible Markdown with strict YAML frontmatter, stored in a Git-managed vault, and synced via standard `git pull/push/commit`. No external API keys, no external service, no network calls.
+
+### Architecture (three-layer model)
+
+```
+   ┌─────────────────────────────────────────────┐
+   │ Layer 1: Markdown is truth                  │  ← canonical store
+   │   .obsidian/  or  ~/.local/share/bizar/…    │     (frontmatter + body)
+   ├─────────────────────────────────────────────┤
+   │ Layer 2: Git is collaboration               │  ← sync + version control
+   │   bizar memory pull / commit / push / sync  │     (branches, conflict)
+   ├─────────────────────────────────────────────┤
+   │ Layer 3: LightRAG is derived index (P2)     │  ← semantic retrieval
+   │   bizar memory reindex                       │     (stub in Phase 1)
+   └─────────────────────────────────────────────┘
+```
+
+The LightRAG layer ships as a stub in Phase 1 (`bizar memory reindex` returns a "not yet implemented" notice) and will be wired into the existing `mods-examples/lightrag/` server in Phase 2. The Markdown layer is the canonical source of truth — reindexing rebuilds the index from Markdown at any time, never the other way around.
+
+### Modes
+
+| Mode | Vault location | Cross-project sharing | Default? |
+|---|---|---|---|
+| `local-only` | `<project>/.obsidian/` | No — per-project vault | **Yes** |
+| `managed` | `~/.local/share/bizar/memory/<repoName>/` | Yes — one shared repo with three namespaces | Opt-in |
+
+Pick `managed` at `bizar init` time, or run `bizar memory link ~/.local/share/bizar/memory/<repoName>/` to convert an existing project to managed mode.
+
+### Namespaces
+
+The `managed` repo is split into three top-level directories. Every note lives in exactly one:
+
+| Namespace | Purpose | Example path |
+|---|---|---|
+| `projects/<projectId>/` | Per-project memory — conventions, ADRs, bugs, commands | `projects/bizarharness/architecture_decision__lightrag-stub.md` |
+| `global/bizar/` | Cross-project knowledge — agent patterns, Bizar internals | `global/bizar/coding_convention__plan-then-forseti.md` |
+| `users/<userId>/` | Personal scratch — user preferences, todos, draft thoughts | `users/drb0rk/user_preference__tabs-not-spaces.md` |
+
+### CLI commands
+
+The `bizar memory <sub>` family has 11 subcommands:
+
+| Subcommand | Purpose | Example |
+|---|---|---|
+| `init` | Create `.bizar/memory.json` and the per-project vault | `bizar memory init --memory-mode local-only` |
+| `status` | Show mode, link target, dirty files, last sync | `bizar memory status` |
+| `link <path>` | Bind the project to a managed memory repo | `bizar memory link ~/.local/share/bizar/memory/work/` |
+| `unlink` | Detach from managed mode (vault stays on disk) | `bizar memory unlink` |
+| `pull` | `git pull` the linked memory repo | `bizar memory pull` |
+| `commit` | Stage dirty notes + run secret scan + `git commit` | `bizar memory commit -m "add auth ADR"` |
+| `push` | `git push` the linked memory repo | `bizar memory push` |
+| `sync` | pull → reindex → commit → push (the common path) | `bizar memory sync` |
+| `reindex` | Rebuild the derived LightRAG index (Phase 2 stub) | `bizar memory reindex` |
+| `conflicts` | List notes with merge conflicts awaiting resolution | `bizar memory conflicts` |
+| `doctor` | Run schema + secrets + Git health checks | `bizar memory doctor` |
+
+### Dashboard routes
+
+The dashboard exposes 18 REST endpoints under `/api/memory/*` for note CRUD, schema validation, secret scanning, Git sync, search, and health checks. The legacy `/api/obsidian/*` routes are preserved with back-compat response shapes.
+
+### Secrets and schema
+
+Every note passing through `bizar memory sync` is scanned against 12 secret patterns (HIGH/MEDIUM). HIGH-severity matches block the commit; MEDIUM matches warn but allow. Required frontmatter is 8 fields (`memory_id`, `type`, `project_id`, `status`, `confidence`, `created`, `updated`, `tags`); 11 memory types are recognized, 6 statuses, 3 confidence levels.
+
 ---
 
 ## 🔑 Provider Setup

package/bizar-dash/.bizar/activity.log

@@ -0,0 +1,3 @@
+{"ts":"2026-06-19T08:39:29.113Z","kind":"settings.update"}
+{"ts":"2026-06-19T08:39:37.682Z","kind":"settings.update"}
+{"ts":"2026-06-19T08:39:44.191Z","kind":"settings.update"}