Loose agents
- Repeated prompts for every task
- Context lost between tools
- You review, forward and stitch everything together by hand
- A single model, a single point of failure
On-demand engineering capacity
From loose agents to an autonomous team.
Ask for it in Slack, it reaches production. ciloops is a team of agents that takes your request, bug or change, turns it into issues and PRs, reviews it against the project, audits it and deploys it. No micromanagement: you test it and reply in the thread.
- Integrated with
- Anthropic
- OpenAI
#product
M
Mariana (PM)
Customers on the annual plan can't see the discount at checkout. Can you take a look?
L
Lena
Found 2 affected repos: billing-api and web-checkout. Opened issues #418 and #92, linked.
B
Bruno
PR ready in both repos. Fix for the proration calculation and a regression test.
V
Vera · Sam
Matches the pricing rule in Notion. No new dependencies or findings. Approve ×2.
D
Diego
Deployed to pre-production. Mariana, can you try it?
M
Mariana (PM)
It works. Ship it to production.
D
Diego
Deployed to production. PRs merged and issues closed, nothing left open. Closing the request.
The problem
A loose agent is a tool. A team is a system.
Each agent waits for your instructions, loses context and doesn't know what the one next to it did. You end up being the coordinator.
ciloops autonomous team
- One request; the team builds the plan within a defined work framework
- Your business context shared across all agents
- Business and security review before deploying
- The model from Google, Anthropic or OpenAI of your choice
They're not agents with good prompts. They're agents with a work framework and a clear picture of your business.
Just as the cloud gave you servers on demand, ciloops gives you engineering capacity on demand: you ask for it from Slack, it's metered per task and it scales with every project.
The loop
Five agents, one request that goes all the way around
Each agent does one thing well and leaves a trail in the repo and the thread. The loop advances on its own; tap a step or an agent to jump straight to it.
Productionthe loop's destination
Step 1 · Channel
You CEO · PM · Architect · Dev
You notify via Slack
A request, a change or a bug. In plain language, in your usual channel. No tool switching.
Mariana (PM)
Customers on the annual plan can't see the discount at checkout. Can you take a look?
- Tools
- Slack
- Deliverable
- A message in the channel
Step 2 · Triage
Lena Triage
Turns the request into issues
Finds which repos are affected, opens one issue for each and links them together.
Lena
Found 2 affected repos: billing-api and web-checkout. Opened issues #418 and #92, linked.
- Tools
- GitHubSlack
- Deliverable
- Linked issues
Step 3 · Build
Bruno Builder
Writes the fix and opens the PR
Implements the change in each repo, adds a regression test and opens the PRs. Two builders are two issues in parallel.
Bruno
PR ready in both repos. Fix for the proration calculation and a regression test.
- Tools
- GitHubYour model
- Deliverable
- PRs with tests
Step 4 · Big picture
Vera Big picture
Reviews against the project
Checks the change against your business rules in your documentation: Notion, Google Drive or Markdown in the repo.
Vera
Matches the pricing rule in Notion.
- Tools
- NotionGoogle DriveMarkdown
- Deliverable
- Business review
Step 5 · Security
Sam Security
Audits the change
Reviews dependencies and security findings before anything moves forward.
Sam
No new dependencies or findings.
- Tools
- GitHub
- Deliverable
- Security audit
Step 6 · Approval
You Decide
Two approves, always
Business and security approve every PR, on every iteration. If something fails, the loop keeps going on its own until you give the OK.
Vera · Sam
Approve ×2.
- Tools
- GitHubSlack
- Deliverable
- Approved PR
Step 7 · Release
Diego Release
Deploys and closes everything
With both approves, it moves the PR to pre-production and tells you in the thread so you can test it. When you reply, it takes it to production and, whatever you decide, merges the PRs, closes the issues and makes sure nothing is left open.
Diego
Deployed to pre-production. Mariana, can you try it?
- Tools
- GitHubSlack
- Deliverable
- Deployed and closed, nothing pending
Integrations
Lives where your team already lives
ciloops isn't another tool to open. It reads and writes in the ones you already use: the channel, the repos, the documentation and the model you chose.
- Anthropic
- OpenAI
Slack
Entry and notification channel. Installs with one button; everything happens in threads.
GitHub
An app that only sees the repos you choose. Issues, PRs and comments.
Notion
Where the business truth lives: pages the agents read.
Google Drive
Product documentation folders, with read-only permission.
Markdown in the repo
A docs folder inside a repository counts too.
Your model
Anthropic, OpenAI or Google, with each project's key.
Coming soon
- Channels
- TeamsDiscordTelegram
- Repositories
- GitLabBitbucket
Control
Your key, your cost, your rules
Agents work with what you configure and nothing more. Every task records what it consumed.
The security of your projects and your keys is our priority every time we make a product decision.
Your own model key
Each project uses the Anthropic, OpenAI or Google key you load, and you choose the model from what that key reaches.
What each task spends
Each task records the tokens it used and what they would cost on your key, next to the compute.
Tokens under control
Agents work with an optimized context to avoid excessive or runaway consumption on every iteration.
Measured compute, not estimated
Each task runs in its own container. You're billed for the vCPU and memory it measured, at the price fixed in the registry.
Conventions you write once
How your team builds, its security bar and where the business lives. Each agent reads only what concerns it.
Sealed secrets
Your keys are kept protected and agents have no access to them.
Scale without switching accounts
Manage all your projects from a single account, each one separately: its repos, its channels and its keys. No connection is shared between projects unless you load it.
Why ciloops
Less context switching.
More deliveries that land.
channel for everything
You ask, follow and approve from Slack. Nobody jumps between six tools.
approves before deploying
Business and security, always. On every PR, on every iteration.
manual handoffs
From message to PR and from PR to pre-production, without anyone having to pass it along.
iterations until it's right
If something fails in testing, the loop keeps going on its own until you give the OK.
FAQ
Frequently asked questions
What's the difference between loose agents and an autonomous team?
A loose agent does a task when you ask and you coordinate the rest. In ciloops each agent has a role, shares context with the others and the loop advances on its own from request to production.
Which model providers does it work with?
It's integrated with Google, Anthropic and OpenAI. Each project uses the key you load, and you choose the model from what that key reaches.
Do I have to switch tools?
No. The request comes in through Slack and the agents work on your GitHub repos and your documentation in Notion, Google Drive or Markdown in the repo.
Can I manage more than one project?
Yes. ciloops scales with you: you manage several projects from the same account, each one separately with its own repos, channels and keys, without switching accounts.
Who approves the changes?
You do. Every PR requires two approves, one for business and one for security, before deploying.
Early access
Connect a channel and a repo. Send the first request.
We're opening small cohorts for teams of 5 to 50 people. We create your account, you connect Slack and GitHub with a button and load your model key.
I want accessWrite to us at hola@ciloops.com
ciloops is a self-service product, but we'll walk you through the process.