fleet · 6 agents · 4 running · 0 failed (24h)
Somewhere for your agents to live once they leave your laptop
Deploy agents built with LangChain, CrewAI, the OpenAI Agents SDK or your own code. We give them a schedule, a secret store, retries when a call fails, and a record of everything they did.
runs today
1,182
failed (24h)
0
throughput
The operations layer agents are missing
You have written the agent. This is the part between it working once and it working every day.
Bring your own framework
If it runs in Python or Node, it runs here. We do not ask you to rewrite your agent into a visual builder or a proprietary DSL to deploy it.
Schedules, webhooks and queues
Run on a cron expression, when an HTTP request arrives, or when a job lands on a queue. One agent can have several triggers without changing its code.
Secrets that are not in your repo
API keys and tokens live encrypted and are injected as environment variables at run time. Rotate one centrally instead of redeploying six agents.
Retries with exponential backoff
Third-party APIs fail. Runs retry on a backoff you set, and only alert you once they have genuinely given up rather than on the first blip.
Cost and duration per run
Every run records tokens, spend, duration and outcome, so you can see which agent is expensive and which one quietly doubled its runtime last week.
Staging that matches production
A separate environment with its own secrets and schedules, so a change can be proved before it runs against live systems.
the options
Three places agents end up
Almost every agent starts on a developer's machine. What happens next usually decides whether it becomes something the business relies on or something that quietly stopped in March.
Framework
- A cron job on a VPS
- Anything
- A no-code agent platform
- Theirs, and only theirs
- The Email Shop
- Anything — LangChain, CrewAI, OpenAI SDK, plain Python or Node
When a run fails
- A cron job on a VPS
- It fails silently unless you built alerting
- A no-code agent platform
- Retried, opaquely
- The Email Shop
- Retried with backoff, then alerted with the full trace
Secrets
- A cron job on a VPS
- Environment variables in a file
- A no-code agent platform
- Their vault
- The Email Shop
- Encrypted store, injected at run time, rotatable
Scheduling
- A cron job on a VPS
- crontab, if you remember the syntax
- A no-code agent platform
- Built in
- The Email Shop
- Schedules, webhooks and queue triggers, all in one place
Run history
- A cron job on a VPS
- Log files, until the disk fills
- A no-code agent platform
- Kept, in their dashboard
- The Email Shop
- Every run: inputs, steps, outputs, cost and duration
Cost per month
- A cron job on a VPS
- £5–£16 plus everything you build around it
- A no-code agent platform
- Roughly £24–£87 for SMB tiers
- The Email Shop
- From £19.99, all in
For a single agent that runs once a day and does not matter much if it misses a day, a cron job on a small VPS is honestly fine.
in production
What changes when an agent becomes load-bearing
The gap between a demo and something a colleague depends on is mostly these four things.
Silent failure is the default failure
An agent that stops working rarely announces it. The cron job still fires, the process still exits zero, and the work simply stops happening — often noticed weeks later when someone asks why the report stopped arriving. Every run here is recorded with an outcome, missed schedules are flagged, and repeated failures raise an alert with the trace attached. The point is to find out on the day, not in the quarterly review.
Retrying badly is worse than not retrying
Agents talk to APIs that rate-limit, time out and occasionally return nonsense. A naive retry loop hammers a struggling service and turns a blip into an outage — sometimes yours, sometimes theirs. Retries use exponential backoff with a ceiling and a maximum attempt count you set per agent, and a run that exhausts them stops and reports rather than continuing to try.
Secrets multiply faster than agents
Six agents, each needing three API keys, is eighteen credentials to place, and every one committed to a repo or pasted into a config file is a small future incident. Secrets live in an encrypted store, are injected at run time, and are referenced by name. Rotating a key is one change in one place, and no agent needs redeploying.
You will want to know what a run cost
Token spend is easy to ignore until an agent's context grows and its cost per run triples without the behaviour visibly changing. Each run records tokens, spend, duration and outcome, so cost is attributable per agent rather than arriving as a single provider invoice you cannot break down.
pricing
Priced by agents, not by runs
An agent running every minute costs the same as one running once a week.
Launch
First agents in production
2 vCPU · 4 GB RAM · 40 GB NVMe · 3 agents
- Any Python or Node framework
- Schedule and webhook triggers
- Encrypted secret store
- +2 more
£19.99/mo
Get startedGrowth
most chosenA working fleet
4 vCPU · 8 GB RAM · 100 GB NVMe · 10 agents
- Everything in Launch
- Ten agents with parallel runs
- Staging environment
- +2 more
£39.99/mo
Get startedScale
Agents across several teams
8 vCPU · 16 GB RAM · 250 GB NVMe · 30 agents
- Everything in Growth
- Thirty agents, isolated per project
- Role-based access for your team
- +2 more
£79.99/mo
Get startedPrices exclude VAT. No setup fee, no minimum term.
Frequently asked questions
Move your agents off the laptop
Deploy from Git, keep your framework, and get scheduling, secrets and run history from day one. From £19.99 a month.