.env.example

# social-post-forge — copy to .env and fill in what you need.
# Nothing here is required for the copy-paste workflow except ANTHROPIC_API_KEY
# when using the CLI. Inside an agent harness you need none of it.

# ---------------------------------------------------------------- CLI only
# Only forge.py needs a model. Running the skill inside an agent harness
# (Hermes, Claude Code, Codex) uses that agent's own model and needs none of this.

# Option A — Anthropic (default)
ANTHROPIC_API_KEY=

# Option B — any OpenAI-compatible endpoint. Set the base URL and the CLI
# switches to /chat/completions automatically.
#   OpenRouter   https://openrouter.ai/api/v1
#   Nous         https://inference-api.nousresearch.com/v1
#   Together     https://api.together.xyz/v1
#   Groq         https://api.groq.com/openai/v1
#   Ollama       http://localhost:11434/v1      (any dummy FORGE_API_KEY)
#   LM Studio    http://localhost:1234/v1       (any dummy FORGE_API_KEY)
# FORGE_BASE_URL=
# FORGE_API_KEY=

# Model override, required for Option B since the default is Anthropic's
# FORGE_MODEL=

# Per-call timeout in seconds. Defaults to 180 hosted, 900 when FORGE_BASE_URL
# is set. A cold quantized model on CPU can exceed both — warm it first with
# `ollama run <model> "hi"` and raise this.
# FORGE_TIMEOUT=3600
# Passes Ollama's keep_alive on every call so the model is not evicted between
# the pipeline's sequential generations. Alternative to OLLAMA_KEEP_ALIVE=30m.
# FORGE_KEEP_ALIVE=30m
# Reasoning models can spend the whole budget thinking and return nothing.
# Set to 0 only if you want thinking tokens enabled.
# FORGE_NO_THINK=1


# ------------------------------------------------ Mode B: self-hosted Postiz
# The backend port, not the 4007 UI port.
POSTIZ_URL=http://localhost:3000
# Postiz UI > Settings > Developers > Public API.
# Sent as a raw Authorization header with no "Bearer " prefix.
POSTIZ_API_KEY=


# ---------------------------------------------------- Mode C: native APIs
# Only fill in the platforms you actually publish to directly.

# LinkedIn — free, no app review needed for w_member_social.
# Access token lasts 60 days and there is no refresh token unless you are a
# Marketing Developer Platform partner, so expect to re-auth every 60 days.
LINKEDIN_ACCESS_TOKEN=
# urn:li:person:XXXX for personal, urn:li:organization:XXXX for a company page
LINKEDIN_PERSON_URN=

# Instagram — free, no review for your own account, but:
#   - account must be Professional (Business or Creator)
#   - linked to a Facebook Page (Facebook Login path)
#   - images must be public HTTPS JPEGs; there is no text-only feed post
# Long-lived token lasts 60 days and is refreshable before expiry.
IG_ACCESS_TOKEN=
IG_USER_ID=

# Threads — free, no review if the posting account is a Tester on your app.
# The easiest of the four to set up.
THREADS_ACCESS_TOKEN=
THREADS_USER_ID=

# X — PAID. Pay-per-use since February 2026: roughly $0.01 per post and about
# $0.20 for a post containing a URL. No free tier for new developers.
# Needs an OAuth 2.0 user token with tweet.write; app-only tokens cannot post.
# The app must be in the production environment or writes return 403.
# Access tokens last 2 hours, so you need refresh logic (offline.access scope).
X_ACCESS_TOKEN=

# TikTok — free, but an unaudited app can only post SELF_ONLY (private).
# Public posting requires passing TikTok's audit, and their guidelines
# explicitly disfavour tools built for accounts you manage yourself.
# Access token lasts 24 hours; refresh token lasts 365 days.
# PULL_FROM_URL requires verifying the domain in the developer dashboard.
TIKTOK_ACCESS_TOKEN=