DaveCode v0.1.0

First release: every phase of the specification is implemented.

Added

• pnpm monorepo with core, server and cli packages, strict TypeScript, Biome and Vitest.
• Shared domain contracts, ProviderError failover classification and the typed EventBus.
• zod configuration schema with ToS-sensitive features disabled by default.
• Architecture, HTTP API and specification documents.
• Dual-brain context system: task graph DAG resolver with cycle paths and status transitions,
  ProjectBrain (STATE.md section helpers, atomic writes, lock file, task.updated events),
  GlobalBrain markdown notes, buildTaskContext with budgeted truncation and the master
  DAVECODE_SYSTEM_PROMPT.
• Provider adapters in @davecode/core: openai, openai-compatible, anthropic, gemini,
  claude-cli (per-account CLAUDE_CONFIG_DIR), codex-cli (per-account CODEX_HOME) and an
  experimental driver-injected gemini-web, all translating to the OpenAI wire format with
  uniform ProviderError mapping, plus the createProviders() registry.
• MIT license, contributing guide, code of conduct, security policy and CI.
• Layered config loader (defaults, global, project, DAVECODE_* env) with file-specific errors.
• SQLite storage (WAL, versioned migrations) with account, usage and audit repositories.
• AES-256-GCM keyring for account secrets with a per-install master key.
• Per-account sandboxes for CLI providers (CLAUDE_CONFIG_DIR, CODEX_HOME) and isolated
  Chromium profiles for experimental browser providers (playwright-core optional).
• Token estimation with js-tiktoken (o200k_base) and a chars/4 fallback.
• Sliding-window quota engine (1 m, 5 h, 24 h) and a SQLite-backed usage tracker.
• Router with quota-aware weighted balancing, backpressure, 5 h quota shift, per-account
  circuit breaker and hot failover (streams fail over only before the first chunk).
• createEngine() composition root wiring config, storage, identity, quotas and the router.
• Fastify gateway: OpenAI-compatible /v1 (JSON + SSE), the full /api surface, live
  /api/events, bearer auth, CORS for the Vite dev server and optional dashboard hosting.
• @davecode/ui web dashboard (Vite, React 19, Tailwind CSS v4): overview with live token burn
  and quota pressure, accounts (add/edit with write-only secrets, gemini-web Terms-of-Service
  gate), traffic with inline failover chains, routes, task graph DAG/board/list with the project
  brain, runner control with a live log console, settings, command palette and keyboard
  shortcuts. Dark by default with a light theme.
• Typed client for every /api endpoint plus an /api/events stream with backoff reconnects,
  and a realistic in-browser mock mode (?mock=1, VITE_DAVECODE_MOCK=1, or automatic when the
  gateway is unreachable).
• Autonomous runner (AutonomousRunner, createRunner): selects the next ready task, works on
  an isolated davecode/task-<id> branch, validates with the configured lint/typecheck/test
  commands, self-repairs up to runner.maxRepairCycles with the exact failure output, runs an
  optional judge, then commits (Conventional Commits) and merges --no-ff or opens a PR with
  gh, updating TASK_GRAPH.json and STATE.md. 24/7 mode with idle polling or runOnce();
  refuses a dirty tree; pause/stop at step boundaries with AbortSignal cancellation.
• Executors: the default builtin OpenAI tool-calling loop over the router (list_dir,
  read_file, write_file, edit_file, search, run_command, finish) with repo path
  confinement and a command allow-list, and opt-in claude-cli delegation using the account's
  sandboxed CLAUDE_CONFIG_DIR.
• Judges: none (default), TypeSafe Jev via the OpenRouter Decisions API (key from
  OPENROUTER_API_KEY / JEV_API_KEY) and llm over your own routes with a zod-validated
  verdict.
• ProjectBrainSource adapter for the gateway; pnpm dev exposes the project brain and runner.
• runner.* config: executor, pullRequests, commitBrain, commandTimeoutMs,
  idlePollMs, maxIterations, maxTaskTokens, allowedCommands and claudeCli.
• davecode CLI (commander, lazy-loaded commands): start (gateway, dashboard, project brain
  and autonomous runner, with a compact banner, live activity log and clean SIGINT/SIGTERM
  shutdown), status (live from /api/* or local state, 1m/5h/24h quota bars), accounts list|add|remove|enable|disable|login plus add-account (interactive Ink prompts with masked
  secrets or a flag form for scripts; gemini-web and multi-account rotation gated behind their
  experimental flags with the ToS warning; login runs the real claude/codex in the
  account's isolated config dir), init (brain scaffold and runner.validate detected from
  package.json), tasks list|next|add|status, config path|show|get (secrets redacted),
  doctor (environment checks with fixes) and run [--once] [--task <id>]. Global --home,
  --json, --no-color/NO_COLOR and --version.
• Chat TUI (davecode chat, or bare davecode in a terminal) built with Ink 8: streaming
  replies over SSE, multiline input, slash commands (/model, /route, /accounts, /tasks,
  /status, /clear, /help, /exit), Esc to cancel, Ctrl+C twice to exit and a status line
  with the serving account, failovers, tokens and 5 h quota; starts an in-process gateway when
  none is running.
• Live runner view for davecode run: state-machine position, current task, repair cycle
  x/max and a log tail, fed by the in-process event bus or a running gateway's /api/events.

Getting started

git clone https://github.com/DMRLZZ/DaveCode.git && cd DaveCode
pnpm install && pnpm build
node packages/cli/bin/davecode.js doctor
node packages/cli/bin/davecode.js start   # gateway + dashboard on :4040

ToS-sensitive features (gemini-web, multi-account rotation) ship disabled. See the README.