Tide chart
Keep AI spend at low tide.
- Spent today
- Daily limit (top row)
- Where today's pace ends up
Today in plain English
Set up your first agent
Three steps. Takes about a minute.
- It can never spend more than this in a day.
Paste these two lines into your agent
Your agent keeps working exactly the same. It just talks to Lowtides first.
Show me code for my agent
Your agents
Try it
Ask anything through one of your agents and see which model answered, what it cost and what it saved.
What the governor noticed
Waiting for you
Checks for new requests every 3 seconds.
Your agents ask Lowtides before they buy anything. Purchases your rules can't settle on their own wait here until you approve or deny them.
Decision log
Every decision, with the reasons Lowtides gave. Each entry is written to a tamper-evident audit log.
Get a price before the work starts
Like a contractor's estimate. Lowtides quotes the job from what similar jobs really cost. Hold the quote on PayPal, and when the job is done you pay only the real cost. The rest is released.
Jobs
Agents add calls to a job by sending the header X-Lowtides-Task: job-id. Charges round to the nearest cent, and anything under half a cent is free.
Prepaid allowance
Turn on "Pay from wallet" for an agent and it spends only what you've loaded. When the wallet is empty, it stops. No surprise bills.
Bill clients for the AI you run for them
Give each client their own agents. Lowtides totals what their agents cost, adds your markup, and sends a PayPal invoice with the savings receipt on it.
Invoices
Every call, every cent
"Would have cost" is the same tokens on the model your agent asked for, with no trimming, no added caching and no answer memory. Cheap attempts that had to be redone are included in what you paid.
Merchants your agents may pay
Paying a merchant that isn't on this list needs your approval, so a prompt-injected agent can't send money to a stranger's PayPal account.
Add a merchant
Rules of the road
Models and prices
Prices are per 1 million tokens. Check them against your provider's pricing page: Lowtides's dollar math is only as right as these numbers.
Add or update a model
Connections
Add an OpenAI-compatible server
Works with OpenAI, OpenRouter, Groq, Together, or a local server like Ollama or LM Studio. Keys live in environment variables, never in this page.
Answer memory
Repeat questions are answered free from memory for 24 hours. Answers that call tools are never reused.