·2 min read·Ryvos Team

Ryvos v0.5.0: Browser Automation, WhatsApp, Budget, and Embeddings

Ryvos v0.5.0: Browser Automation, WhatsApp, Budget, and Embeddings

Today we're shipping Ryvos v0.5.0 — the most feature-packed release since v0.2.0. Here's what's new.

Browser Automation (5 New Tools)

Ryvos can now control a browser. Five new T3 tools powered by Chromium give your agent the ability to navigate websites, take screenshots, click elements, type into forms, and extract content:

ToolDescription
browser_navigateNavigate to a URL
browser_screenshotScreenshot the current page
browser_clickClick an element by CSS selector
browser_typeType text into an input field
browser_extractExtract text content from the page

All browser tools are classified at T3 and the agent applies constitutional reasoning before acting. See the Built-in Tools docs.

WhatsApp Cloud API

Ryvos now supports WhatsApp as a channel. Connect your Meta Business app, configure your webhook, and chat with your agent from WhatsApp — including interactive approval buttons for tool calls.

[channels.whatsapp]
access_token = "${WHATSAPP_ACCESS_TOKEN}"
phone_number_id = "${WHATSAPP_PHONE_NUMBER_ID}"
verify_token = "${WHATSAPP_VERIFY_TOKEN}"
allowed_users = ["+1234567890"]
dm_policy = "allowlist"

See the Slack & WhatsApp docs.

18+ LLM Providers

Three new providers bring the total to 18+:

  • AWS Bedrock — full SigV4 authentication support for Claude, Llama, and other models on AWS
  • Claude Code — delegate to the Claude Code CLI, using your Anthropic subscription billing
  • GitHub Copilot — delegate to the Copilot CLI, requires a GitHub Copilot license

See the Providers guide.

Budget System

Track and limit your LLM spending with monthly budgets:

[budget]
monthly_budget_cents = 2000       # $20.00/month
warn_pct = 80
hard_stop_pct = 100

The Guardian monitors usage after each LLM call and publishes BudgetWarning / BudgetExceeded events. Notifications route to your configured channels. See the Configuration docs.

Semantic Memory (Embeddings)

Configure an embedding provider to upgrade memory_search from BM25 (keyword) to hybrid vector similarity:

[embedding]
provider = "openai"
model = "text-embedding-3-small"
dimensions = 1536
api_key = "${OPENAI_API_KEY}"

Supports OpenAI, Ollama (local), and any OpenAI-compatible API. See the Memory docs.

10-Phase Onboarding

ryvos init now walks through a comprehensive 10-phase setup:

  1. Provider selection
  2. Model configuration
  3. Security tier
  4. Channel setup
  5. Sandbox configuration
  6. Guardian watchdog
  7. Heartbeat
  8. Budget limits
  9. Embedding provider
  10. Summary and confirmation

Each phase has sensible defaults — press Enter to skip anything you don't need.

6 Bundled Skills

v0.5.0 ships with 6 built-in skills ready to use out of the box:

  • web-scraper — extract structured data from websites
  • code-review — automated code review with security checks
  • git-summary — generate changelogs from git history
  • docker-deploy — build and deploy Docker containers
  • db-migrate — run database migrations safely
  • api-test — test REST API endpoints

Install community skills with ryvos skill install <name> or list bundled ones with ryvos skill list.

Upgrading

cargo install --path . --force
# or
curl -fsSL https://raw.githubusercontent.com/Ryvos/ryvos/main/install.sh | sh

Your existing config is fully backward-compatible. All new config sections ([budget], [embedding], [channels.whatsapp]) are optional.


Full changelog on GitHub.