Kiro bots are now generally available in AWS’s Kiro Crew workspace. A bot is a named AI agent with its own memory and schedule, and it can run overnight checks without you, as VentureBeat reported after the November 5 announcement.
The catch is the bill. Bot runs draw from the same credit pool as your normal Kiro work, so a bot that checks something every 30 minutes adds up fast.
Here are the facts most readers want first:
- Announced and made generally available on November 5, 2026.
- Bots live inside Kiro Crew and can run on a laptop, a Mac Mini or an Amazon EC2 instance.
- Team plans cost $20 per seat per month with 500 pooled credits per seat.
- Extra credits cost $0.04 each, and there is no separate per-bot fee.
- Memory is kept per bot, and you can view, edit or delete it.
Everything below explains what those facts mean in practice.
What Kiro bots actually do
Each bot has a name, a continuing conversation, its own notes and a dedicated workspace. You tell it what job to own, and it can start several worker sessions, track their progress and pull the results together. If one task needs your decision, the bot asks while the others keep going.
You set schedules in plain chat. Think “investigate CI failures overnight” or “send me a weekly summary of merged changes.” An activity log records the routine work, and the chat only surfaces what needs your attention.
This is the same shift we covered when Cursor started hosting apps for you: coding tools are moving from “answer when asked” to “keep working on their own.”
Memory and permissions: what stays in your control
Memory belongs to one bot and is not shared with the others. You can inspect it, correct it or wipe it. Per the Beri.net analysis, lessons are stored as Markdown files with a signed audit log.
Permissions are set per bot, and adding a new tool needs your approval. AWS gives these examples of how narrow a bot can be:
- A research bot can read files but not edit them.
- A deployment bot asks before every change.
- A review bot can comment and request changes, but it cannot approve or merge code.
That last limit matters. A bot that can only suggest is much safer to leave running while you sleep.
One risk is not addressed in the launch. Beri.net points out that a bot reading a public issue tracker could pick up misleading text and save it as a “lesson.” Keep bots that read untrusted input on tight permissions.
What Kiro bots cost: the credit math
Kiro subscriptions cover Crew, the IDE, the CLI and the web product, so bots do not add a flat fee. Instead, every bot run spends credits, and the price of a credit depends on which model does the work. Beri.net listed these multipliers as of November 6:
| Model | Credit multiplier |
|---|---|
| Claude Fable 5.1 | 6x (Enterprise only, admin must enable) |
| GPT-5.6 Sol | 4.4x (8.8x past 272K tokens) |
| Claude Opus 5.5 | 2.0x |
| Claude Sonnet 5.5 | 1.3x |
| GPT-5.6 Luna | 0.6x |
| Claude Haiku 5.5 | 0.06x (0.3x past 100K tokens) |
Bots can use one model to plan and a cheaper one for workers. By default, workers use the same model as the bot, so you need to change that setting yourself.
Here is a hypothetical example from the same analysis, since Kiro does not publish a per-check cost. A 10-seat team has a 5,000-credit pool. A build-health bot checking every 30 minutes runs about 1,440 checks a month. At one credit each, that is roughly 29% of the pool, or $57.60 at the add-on price.
Scheduled jobs that only run a plain script, with no model call, use no credits. Admins can set usage limits per person or group, with alerts at 80% and 100%.
Four ready-made bots worth trying first
AWS ships reference bots that show the intended pattern. Each one has a clear job and a clear limit:
| Bot | What it does | What it cannot do |
|---|---|---|
| Intake bot | Labels tickets, flags duplicates, drafts reproduction steps | Close tickets |
| Build-health bot | Checks CI every 30 minutes and separates new failures from known flaky tests | Change code on its own |
| Code review bot | Checks pull requests against past decisions | Approve or merge |
| Release-readiness bot | Checks tests and release notes before shipping | Ship without your sign-off |
If you already run agents elsewhere, compare usage rules first. We looked at how Claude Code subagent usage counts against limits, and the same question applies here.
Gaps AWS has not filled yet
The launch post has no benchmarks, error rates or productivity numbers. It also does not say how much you save by pairing a strong planner with cheap workers.
Another open question is whether bot usage counts against the creator’s personal limit or only the shared pool. Beri.net recommends asking your AWS account team in writing before you roll bots out widely.
Also note that the dashboard binds to localhost:5476 and should not be exposed to the internet. Dynamic capabilities, where a bot asks for new tools through a chat card, is still in preview.
Frequently Asked Questions
Are Kiro bots free?
Not entirely. The free plan includes 50 credits, and bots spend credits like any other Kiro work. There is no separate per-bot fee.
Where do Kiro bots run?
On your laptop, a Mac Mini or an Amazon EC2 instance. Kiro Crew is open source under Apache 2.0 and runs on the Kiro CLI.
Can a Kiro bot merge code by itself?
Only if you allow it. The reference review bot can comment and request changes but cannot approve or merge.
Does a bot learn between runs?
It keeps saved notes and corrections in its own memory. The launch post does not say the underlying model is retrained.
Can I see what a bot remembers?
Yes. You can view, edit or delete a bot’s memory at any time.
Kiro bots are a good fit if you have boring, repeatable chores like triaging issues or watching CI. Start with one read-only bot, set a credit limit and cheap worker model on day one, and check the usage report after a week. Skip the always-on schedules until you know what a single check costs, because that number is the one AWS has not given you.
Leave a Reply