OpenLivery v0.4.0

The WhatsApp QR bridge rewritten in Go on whatsmeow (sessions survive restarts; voice notes fixed), WhatsApp gestures on both channels for the AI and for operators (read receipts, typing, reactions, quoted replies, incoming gestures), contact merge and country-aware phone entry, and portal fixes.

Upgrading: database migrations apply automatically with the Docker stack. The bridge is a new service and existing Baileys sessions cannot migrate, so each connected QR channel scans the QR once after upgrading. The bridge needs a session store (WHATSAPP_STORE_URL); the bundled compose points it at the stack's Postgres automatically.

See CHANGELOG.md for the full list.

OpenLivery v0.3.0

WhatsApp Cloud API channel (official Meta API)

New channel type, independent from the QR channel: a client can have both connected on different numbers.

Added

• Bring your own Meta app: phone number ID, optional WABA id, permanent access token and app secret. Secrets are encrypted at rest and write-only (never returned by the API).
• Per-channel webhook with the Meta verify handshake and HMAC-SHA256 signature validation over the raw request body. The UI shows the callback URL and verify token with copy buttons and the Meta setup steps.
• Inbound text, image and audio. Media is downloaded from the Graph API and goes through the same transcription/description pipeline as the QR channel; Meta webhook retries are deduplicated by message id.
• AI replies and human operator replies (Inbox and client portal) are delivered through the Graph API.
• Connect and verify validates the credentials against the Graph API and captures the number and verified name.
• New "WhatsApp API" card on the Channels page and on the client channels tab, plus an inbox filter and badge for the new channel.
• META_GRAPH_BASE_URL setting (optional) to pin a Graph API version or point the channel at a mock server.

Changed

• The Baileys channel is now labeled "WhatsApp QR" across the UI to distinguish it from the official API channel.
• The inbound WhatsApp pipeline (dedupe, conversation lookup, media handling, human takeover, AI reply) is shared between both channels.

Upgrading

This release adds a database migration, applied automatically by the Docker stack; run alembic upgrade head on local setups.

v0.2.0

Custom tools for agents: HTTP endpoints and MCP servers

Agents can now call external tools during a conversation, configured per agent from a new Tools tab:

• HTTP tools: user-defined endpoint with {param} path placeholders, method, body and query parameters, prompt instructions, optional auth headers (encrypted at rest) and timeout.
• MCP servers: connect external MCP servers over SSE or Streamable HTTP, with optional auth headers. Connections are tested (tools listed) before saving, and the discovered tool list is cached so chat replies never block on discovery.
• Tool-calling loop for both providers (OpenAI Responses API and Anthropic Messages API). Tool usage is recorded on each reply and shown in the playground, including error details when a call fails; failing tools make the agent report the information as unavailable instead of answering from memory.
• SSRF guard for HTTP tools (private, loopback and reserved addresses rejected; redirects never followed), with a TOOLS_ALLOW_PRIVATE_URLS opt-out for self-hosted deployments.

Also in this release

• Unread count and last-message preview on inbox conversations.
• Channel badge on inbox conversations.
• Fixed spacing of stacked provider cards in settings.
• Full documentation site at openlivery.com/docs, with the README restructured around it.

Upgrading

• A new database migration is included. The Docker stack applies it automatically on start; local setups must run alembic upgrade head.
• The backend has a new dependency (mcp): run pip install -r requirements.txt on local setups.

Full changelog: v0.1.0...v0.2.0

v0.1.0

First tagged release of OpenLivery — a multi-tenant platform for agencies to build and manage AI agents for their clients, with a chat playground, client portal and WhatsApp integration.

Highlights

• Multi-tenant, agency-scoped platform (agencies → clients → agents → conversations).
• FastAPI backend (JWT cookie auth, SQLAlchemy, Alembic) + Next.js web app + Baileys WhatsApp bridge.
• AI chat over any OpenAI-compatible endpoint; OpenAI and Anthropic model presets (incl. the GPT-5.6 family and gpt-transcribe).
• Knowledge documents with semantic retrieval; structured agent business brief.
• Typed i18n (English/Spanish); embeddable chat widget; client portal with custom domains + on-demand TLS.
• Docker Compose + Makefile, single-origin Caddy gateway, prebuilt GHCR images, rate limiting.

See CHANGELOG.md for the full list.