Devlog

A running changelog of meaningful updates — written as we ship.

2026-03-18 — Custom endpoints & multi-backend support (v0.1.8-alpha)

Pro+ users can now point Billy at any OpenAI-compatible server — Groq, OpenRouter, LM Studio, or your own.

Custom backend configuration

[backend]
type    = "custom"
url     = "https://api.groq.com/openai/v1"
model   = "llama-3.3-70b-versatile"
api_key = "gsk_..."

Or use environment variables with no config file at all:

BILLY_BACKEND_TYPE=custom \
BILLY_BACKEND_URL=https://api.groq.com/openai/v1 \
BILLY_BACKEND_MODEL=llama-3.3-70b-versatile \
BILLY_API_KEY=gsk_... \
billy

`/backend` command

Type /backend to see your active backend, model, and the config file path. /backend reload hot-reloads your config without restarting Billy.

Ollama is still the default — free tier and local mode work exactly as before. Custom endpoints require a Pro or higher license.

2026-03-18 — Licensing polish & UX (v0.1.7-alpha)

Polished /activate, /deactivate, and /license prompts — clearer status messages
One-shot mode now respects tier limits consistently with TUI mode
README rewritten — full command reference, config guide, env var overrides, roadmap
LemonSqueezy variant ID map updated for both test and live products

2026-03-16 — LemonSqueezy native activation (v0.1.6-alpha)

Replaced custom Ed25519 key crypto with LemonSqueezy’s built-in License Keys API.

Activation calls POST /v1/licenses/activate — seat-limited, phone-home on first use
/deactivate frees your seat so you can move to another machine
Upgrading a key (e.g. Pro → Premium) auto-deactivates the old seat first
Background re-validation every 7 days keeps licenses current
Activation stored encrypted in SQLite — no plaintext keys on disk

2026-03-16 — Agentic mode, collapsible output & TUI polish (v0.1.1-alpha)

Big batch of TUI improvements land across the board.

Agentic mode

Billy can now propose and run shell commands. When he does, you get an interactive arrow-key picker instead of a y/n prompt:

  ❯  ✅ Approve
     ⏭  Skip
     🚫 Abort
     ▶▶ Run all remaining

After a command runs, the output is fed back to Billy as Command > context — so he can read the result, catch errors, and retry automatically until it works.

Collapsible command output

Long command output (> 15 lines) is collapsed by default with a preview:

$ npm install
  ... 10 lines shown ...
  ╰─ [+] 47 lines hidden · click or Ctrl+X to expand

Billy still receives the full output for reasoning. Click the hint or press Ctrl+X to expand in-place.

Progress bars

Charmbracelet’s harmonica spring-physics progress bars now show during long-running operations (builds, installs, pulls).

Mouse support

Click the [+] N lines hidden hint to expand collapsed output
Ctrl+X keyboard shortcut as an alternative

Label alignment fix

You >, Billy >, and Command > labels are now consistently left-aligned. A lipgloss trailing-newline quirk was causing the first user message to render offset from the left edge. Fixed at the root.

Earlier — Core foundation

Interactive TUI built with Bubble Tea
Local Ollama backend (chat, model list, pull, switch)
Conversation history persisted to SQLite
Memory system — Billy remembers facts you tell him
Session picker — resume any previous conversation
/ command picker with fuzzy dropdown
Homebrew tap, Scoop bucket, .deb/.rpm packages