Reference
Usage limits
Two separate limits apply to assistant usage: a longer-term allowance and a shorter burst window. Knowing which one you hit tells you how long to wait.
Two limits, not one
- Your allowance
- The total you get over a longer period. This is the headline number attached to your plan, and it is what you are really buying.
- The burst window
- A smaller cap over a much shorter rolling period. It exists so one very heavy session cannot consume a whole allowance in an afternoon.
Hitting the window is common and recovers quickly — usually within hours. Hitting the allowance means waiting for the longer reset, or moving up a plan. When you are blocked, the message tells you which one it was and when it comes back.
Current allowances by plan are on the pricing page. They change, and a copy here would go stale.
What actually uses it up
Usage is driven by how much work a request takes, not by how many messages you send. Roughly, from cheapest to most expensive:
| Kind of request | Relative cost |
|---|---|
| A short question with no tool use | Very low |
| Reading a sequence and reporting on it | Low |
| An analysis — cut sites, a digest, primer design | Low, plus a small amount for each external lookup |
| Planning a construct with literature searching | Moderate to high — many steps and much reading |
| Building and verifying a construct | High — the most expensive ordinary operation |
| A GeneLoop | Highest. It runs several phases without checking in. |
You can see where yours has gone in Settings → Usage.
Getting more out of the same allowance
- Put standing constraints in PLASMID.md
- A plan built on wrong assumptions gets redone, and redoing it costs the same again. Project instructions.
- Start a new thread when the topic changes
- A very long conversation carries all of its history into every new message. Threads are grouped by project, so nothing is lost by starting a fresh one.
- Use the bench tools directly for direct questions
- “Does EcoRI cut this?” is free from Enzyme Search and costs usage through the assistant.
- Plan before you build
- Catching a wrong assumption in a plan is much cheaper than catching it in a finished construct.
When you are blocked
You get a clear message rather than a failure: which limit you reached, what remains, when it resets, and what a higher plan would give you. Nothing in progress is lost, and your projects stay open and readable.
If you are using the API
Programmatic access is limited by requests per day, separately from assistant usage.
| Tier | Requests per day |
|---|---|
| Free | 50 |
| Pro | 500 |
| Team | 2,000 |
| Enterprise | Effectively unlimited |
Every response carries your current standing in its headers, so a well-behaved client never has to guess:
| Header | Meaning |
|---|---|
| X-RateLimit-Limit | Your daily allowance. |
| X-RateLimit-Remaining | How many requests are left. |
| X-RateLimit-Reset | When the window resets. |
Exceeding it returns 429 with a Retry-After header. Programmatic access has the details.
Why there is a burst window at all
Because the alternative is worse. Without it, a single runaway session can consume a month's allowance in an afternoon, and the first you would know is finding yourself locked out with three weeks to go. A window that recovers in hours fails much more gently than an allowance that recovers in weeks.