KeyForge

Embeddable API key management SDK with LLM token budget tracking.

from keyforge import KeyForge

kf = KeyForge("sqlite:///keys.db")
await kf.setup()

# Create key — plaintext shown once
result = await kf.create_key(
    name="Production",
    rate_limit={"requests": 100, "window": "1m"},
    token_budget={"monthly": 1_000_000},
    permissions=["models.invoke"],
)
print(result.plaintext)  # kf_live_abc123...

# Verify on every request
v = await kf.verify(result.plaintext)
print(v.valid, v.permissions, v.token_budget_remaining)

# Track LLM usage
await kf.track_usage(result.key.id, input_tokens=500, output_tokens=200, model="claude-sonnet-5-5")

Install

pip install keyforge                    # core (SQLite)
pip install 'keyforge[cli]'             # + CLI
pip install 'keyforge[all]'             # everything

CLI

keyforge init
keyforge create --name "my-key" --rate-requests 100 --monthly-tokens 500000
keyforge verify kf_live_xxx
keyforge list
keyforge usage key_xxx
keyforge rotate key_xxx
keyforge revoke key_xxx

Storage backends

• SQLite (default) — dev / single-server
• PostgreSQL — pip install 'keyforge[postgres]', use postgresql://... URL
• Redis — pip install 'keyforge[redis]', for rate limiting

FastAPI middleware

from keyforge.middleware.fastapi import KeyForgeAuth

auth = KeyForgeAuth(kf, require_permissions=["models.invoke"])

@app.post("/chat", dependencies=[Depends(auth)])
async def chat(request: Request):
    key_id = request.state.key_id  # injected by middleware

MCP server (LLM agents)

keyforge serve

Tools: create_api_key, verify_api_key, list_api_keys, revoke_api_key, track_token_usage, get_usage_stats

Key design

• Format: kf_live_<256-bit-random> (Stripe pattern, enables GitHub secret scanning)
• Hash: SHA-256 (high-entropy keys; argon2 too slow for hot-path verification)
• Rate limiting: sliding window algorithm (Cloudflare-proven)
• Plaintext shown once at creation — never stored