.env.example

# --- Phantom Connect (https://portal.phantom.com) ---------------------------
# Use the React SDK in-app; paste the same App ID here. In Portal add every
# origin under "Allowed Origins" and every callback under "Redirect URLs"
# — run the app once and read values from Bun logs / RPC `phantomGetPortalConfig`
# (`portalAllowedOrigins`, `portalRedirectUrls`) or inspect the handler in
# `src/bun/core/rpc/handlers/phantom.ts`.

PHANTOM_CONNECT_APP_ID=

# Optional: exact redirect URL (must also appear in Portal → Redirect URLs).
# If unset, derived from DETOUR_DEV_URL + portless (see phantom handler).
# PHANTOM_CONNECT_REDIRECT_URL=https://your-domain.example.com/

# Dev shell origin (Vite / tunnel). Non-loopback hosts skip portless and use
# this origin directly as redirect — good for ngrok / Cloudflare Tunnel.
# DETOUR_DEV_URL=http://127.0.0.1:5183

# Public DNS name that resolves to your machine (Portal TXT verification).
# Portless routes this host → DETOUR_DEV_URL port when the proxy is running.
# PHANTOM_PORTLESS_FQDN=phantom-auth.your-domain.example.com

# --- GMGN OpenAPI Agent (https://openapi.gmgn.ai) --------------------------
# 1. Generate an Ed25519 PEM key pair locally (`openssl genpkey -algorithm ed25519`).
# 2. Upload the PUBLIC key at https://gmgn.ai/ai → paste the issued API key here.
# 3. Bind one or more wallets (Solana / BSC / Base / ETH) to the API key in
#    the GMGN dashboard. Those wallets are GMGN-hosted custody and are what
#    `GMGN_SWAP from=...` operates on — NOT the user's Phantom wallet.
# Query endpoints (token/market/portfolio/trending) need only GMGN_API_KEY.
# Trade endpoints (/v1/trade/*) additionally sign each request with
# GMGN_PRIVATE_KEY (the PEM pair to your uploaded public key).
# GMGN_API_KEY=
# GMGN_PRIVATE_KEY=  # PKCS#8 PEM; if pasted on a single line, escape newlines as \n

# --- Dev inference (free OpenAI-compatible proxy) --------------------------
# Routes TEXT_SMALL/MEDIUM/LARGE to a local OpenAI-compatible proxy for dev +
# testing (see src/bun/plugins/dev-inference). Text-only: embeddings and your
# real provider/quota state are untouched. The endpoint, key and models all
# have baked-in defaults (shared rotating free-tier key + the proxy's `auto`
# router) so smoke/integration tests run without any setup — set the flag to
# activate it for the running app.
# DETOUR_DEV_INFERENCE=1
# DETOUR_DEV_INFERENCE_URL=http://localhost:3001/v1
# DETOUR_DEV_INFERENCE_API_KEY=freellmapi-...
# DETOUR_DEV_INFERENCE_MODEL=auto            # one value for all tiers
# DETOUR_DEV_INFERENCE_SMALL_MODEL=auto      # optional per-tier overrides
# DETOUR_DEV_INFERENCE_MEDIUM_MODEL=auto
# DETOUR_DEV_INFERENCE_LARGE_MODEL=auto

# --- Web search (eliza plugin-web-search, Tavily) --------------------------
# When set, the agent gets a fast factual web-search path (the SEARCH "web"
# category) instead of heavyweight browser automation. The plugin is only
# registered when this key is present. Get a key at https://tavily.com.
# Can also be added via Settings → Vault → Inventory (mirrored to env).
# TAVILY_API_KEY=

# --- MCP client (eliza plugin-mcp) -----------------------------------------
# When set, the agent gains tools/resources from the listed MCP servers. JSON:
# either the full { "servers": { … } } shape or just the bare servers map.
# Piped into character.settings.mcp at runtime build. Plugin is only
# registered when this is present.
# MCP_SERVERS={"servers":{"filesystem":{"command":"npx","args":["-y","@modelcontextprotocol/server-filesystem","/Users/me/notes"]}}}