Copied to clipboard

tapes

tapes start

Start tapes and optionally launch an agent. Automatically picks available ports and manages the daemon lifecycle.

Usage

# Start tapes in foreground
tapes start

# Launch an agent through the proxy
tapes start claude
tapes start opencode
tapes start codex

# Launch opencode with a specific provider and model
tapes start opencode --provider anthropic --model claude-sonnet-4-5
tapes start opencode --provider ollama --model qwen3-coder:30b

# Stream logs from the running daemon
tapes start --logs

Supported Agents

Agent Description
claude Launch Claude Code with Anthropic routing
opencode Launch OpenCode with multi-provider routing
codex Launch Codex with OpenAI routing (requires tapes auth openai)

Flags

Flag Description
--logs Stream logs from the running tapes daemon
-d, --debug Enable debug logging
--config-dir Path to tapes config directory
--provider LLM provider for opencode (anthropic, openai, ollama)
--model Model for opencode (e.g. claude-sonnet-4-5, gpt-5.2-codex)
--postgres PostgreSQL connection string (DSN) for persistent storage

All flags can also be set via TAPES_* environment variables — see environment variables. The project name is configurable via TAPES_PROXY_PROJECT or proxy.project in config.toml.

OpenCode Provider Selection

When launching OpenCode, you can specify a provider and model via CLI flags. If no flags are provided:

  1. tapes checks for saved preferences in config.toml
  2. If no saved preference and running in a terminal, prompts interactively
  3. Otherwise, defaults to Anthropic with claude-sonnet-4-5

Your selection is saved to ~/.tapes/config.toml for future sessions.

# First run: interactive prompt
tapes start opencode
# Select a provider for opencode:
# 1) anthropic
# 2) openai
# 3) ollama
# Enter choice [1]: 1
# Enter model [claude-sonnet-4-5]:

# Subsequent runs use saved preference
tapes start opencode

# Override saved preference with flags
tapes start opencode --provider ollama --model codellama:7b

Switching models inside OpenCode won't be captured by tapes. Use separate sessions for different models.

Credential Injection

When launching an agent, tapes automatically injects API credentials stored via tapes auth as environment variables. Shell environment variables take precedence over stored credentials.

# Store your API key once
tapes auth openai

# Keys are auto-injected on every launch
tapes start codex

How It Works

When you run tapes start <agent>:

  1. tapes spawns a background daemon if one isn't already running
  2. The daemon starts proxy and API servers on automatically selected ports
  3. Stored credentials are injected as environment variables
  4. The agent is launched with environment configured to route through the proxy
  5. Conversations are tagged with the agent name for filtering
  6. The daemon shuts down automatically when all agents exit

Use tapes start for the simplest setup. Use tapes serve when you need explicit control over ports and configuration.

Last updated: