Cacheforge Setup

When to use this skill

Use this skill when the user wants to:

• Set up CacheForge for the first time
• Register a new CacheForge account
• Connect their LLM API provider to CacheForge
• Get a CacheForge API key

Setup Flow

• Detect existing API keys — Check for OPENAI_API_KEY, OPENROUTER_API_KEY, ANTHROPIC_API_KEY in the environment
• Collect credentials — Ask user for email and password if not provided
• Auto-detect provider — Infer upstream kind from key prefix:
• sk-or- → openrouter
• sk-ant- → anthropic
• sk- → custom (OpenAI-compatible; legacy openai alias still accepted)
• Preset default base URLs:
• openrouter → https://openrouter.ai/api/v1
• anthropic → https://api.anthropic.com
• custom → https://api.fireworks.ai/inference/v1
• Provision — Run python3 setup.py provision to register/authenticate and get a CacheForge API key
• If registration mode is invite-only, pass --invite-code (or set CACHEFORGE_INVITE_CODE).
• If email verification is enabled, complete verification and rerun provision to mint the tenant API key.
• Validate — Run python3 setup.py validate to make a test request through the proxy
• Configure OpenClaw (recommended) — Print the exact OpenClaw snippet and (with approval) apply it to ~/.openclaw/openclaw.json
• Print: python3 setup.py openclaw-snippet
• Apply: python3 setup.py openclaw-apply --set-default
• If upstream is OpenRouter, the snippet registers multiple popular models so users can switch in /model immediately.
• Fund credits — Before first proxy traffic, top up at least $10 via Stripe or crypto:
• python3 skills/cacheforge-ops/ops.py topup --amount 10 --method stripe
• python3 skills/cacheforge-ops/ops.py topup --amount 10 --method crypto

• Vault Mode virtualizes tool outputs only when the request advertises a fetch-capable tool definition (web_fetch or browser).
• Without a fetch tool definition, CacheForge fail-opens with reason no_fetch_tool.

Commands

# Full setup (interactive)
python3 skills/cacheforge-setup/setup.py provision \
  --email [email protected] \
  --password "..." \
  --invite-code "..." \
  --upstream-kind custom \
  --upstream-base-url https://api.fireworks.ai/inference/v1 \
  --upstream-key fw_...

# Just validate an existing setup
python3 skills/cacheforge-setup/setup.py validate \
  --base-url https://app.anvil-ai.io \
  --api-key cf_...

# Print the OpenClaw snippet (same structure as the CacheForge console)
python3 skills/cacheforge-setup/setup.py openclaw-snippet \
  --base-url https://app.anvil-ai.io \
  --api-key cf_...

# Apply CacheForge provider config into OpenClaw (JSON5-safe; prompts for approval)
python3 skills/cacheforge-setup/setup.py openclaw-apply \
  --base-url https://app.anvil-ai.io \
  --api-key cf_... \
  --set-default

Environment Variables

• CACHEFORGE_BASE_URL — CacheForge API base (default: https://app.anvil-ai.io)
• CACHEFORGE_API_KEY — Existing API key (skip provisioning if set)
• CACHEFORGE_INVITE_CODE — Invite code (required on invite-only deployments)
• OPENAI_API_KEY, OPENROUTER_API_KEY, ANTHROPIC_API_KEY, FIREWORKS_API_KEY — Auto-detected for upstream
• UPSTREAM_BASE_URL — Optional override for provision upstream base URL

After Setup

Once provisioned, set:

export OPENAI_BASE_URL=https://app.anvil-ai.io/v1
export OPENAI_API_KEY=cf_...  # your CacheForge tenant API key

All OpenAI-compatible tools (OpenClaw, Claude Code, Cursor, any agent framework) will route through CacheForge automatically.

If you prefer the OpenClaw-native provider approach (recommended), keep secrets out of openclaw.json and set:

export CACHEFORGE_API_KEY=cf_...

API Contract (current)

This skill uses:

• POST /api/provision
• GET /v1/account/info