Better-AI Dashboard orchestrates the AI coding CLIs you already pay for — plus your own API keys — from one dashboard. Everything runs and stays on your machine.
no account · no telemetry · runs on localhost
One registry · 14 CLIs · auto-detected · one-click install
Features
A real UI on top of headless coding agents — for engineers and beginners alike. Your CLI subscriptions and your API keys are first-class citizens of the same run engine, cost views, and history.
Pick a project, a CLI, a model, a prompt, an optional budget — then watch the live output stream with a running cost. Timeouts, per-run git worktrees, kill escalation, one-click session resume.
Adapters normalize the JSONL session logs your CLIs already write into one SQLite warehouse — no API key needed. 30-day charts, burn-rate gauges, cache efficiency, FTS5 full-text search.
Chain steps with a CLI or API provider, model, and prompt template per step. Ships with a ready-to-run B-MAD template.
Fan one task out to 2–5 CLIs in parallel and let a judge model rank the candidates on a podium.
Automatic detection of installed CLIs, one-click install for the missing ones, per-CLI headless flags and parsers — plus interactive PTY terminal tabs (xterm.js) on any project.
Streaming chat with slash commands from your prompt library, tool-call rendering, and a live token/cost ticker.
Describe what you want in plain language — the Builder generates skills, plugins, workflows, prompts, and agent profiles, and you review everything before it's saved.
A curated catalog of subagents, slash commands, and MCP servers — installed into your setup with one click.
Remotion templates · Hyperframes (prompt → animated HTML composition, browser preview) · text-to-video models via fal.ai (Kling, MiniMax, Veo 2).
Ordered scenes with AI-generated images, dialogue, and durations. Reorder as the story evolves, then export a print-ready contact sheet.
Generate, then iterate: variations on any image, reusable generation presets, and a searchable local gallery with per-image cost.
Every prompt keeps its full history with colored diffs. Run two versions side by side and compare outputs — with the cost of each side.
Recurring schedules with overlap protection and webhook/toast delivery, plus goal-driven loops that iterate until a success criterion is met.
Add, remove, and test MCP servers across scopes — with file backups. Safe editors for CLAUDE.md, AGENTS.md, settings.json, and .mcp.json.
How it works
01
Clone, npm install, run. It auto-detects the CLIs on your machine and the session logs they already write — one click installs the missing ones, or bring your own API keys.
02
Headless runs, multi-step workflows, councils, schedules — each with its own CLI or API provider, model, and budget. Live output streams straight to the dashboard.
03
Everything lands in one SQLite warehouse: costs, burn-rate against a weekly budget, per-model and per-project breakdowns, full-text search across every message.
Under the hood — one app, one warehouse
Sources
Custom Node server
SQLite warehouse
Next.js UI
Launched runs, API-backend runs, and imported CLI history all persist into the same tables — so every cost and history view sees everything. A missing CLI? The API backend runs the same job with the same schema and budget enforcement.
Designed as a local, single-user tool. Core features need zero API keys — and what you do add never leaves your machine.
0
accounts
No sign-up. No cloud.
0
telemetry
Nothing phones home.
localhost
only
The server never leaves your machine.
AES-256
GCM
Keys encrypted at rest, masked last-4.
764
tests
Hermetic — no real CLIs or APIs.
State-changing requests are CSRF-guarded. Process spawning uses argument arrays only (shell:false) — client input never reaches argv unvalidated. Secrets are shown masked (last 4) and always excluded from backups.
Quick start
Requirements: Node ≥ 20.1, npm, git — and at least one AI coding CLI installed (or install one from Settings, or use the API backend with your own keys).
$ git clone https://github.com/supergmax/better-ai-dashboard
$ cd better-ai-dashboard && npm install
$ echo "DASH_MODE=local" > .env.local
$ npm run dev
▲ Ready on http://localhost:3000
Then open localhost:3000 and ingest your CLI history with one request.