A team of AI agents, on call in Slack, GitHub, and Linear.

Ask Echelar to investigate an incident, fix a bug, review a pull request, or update the docs. Its agents do the job in an isolated sandbox, grounded in how your codebase works, and hand the result back where you asked.

Nothing merges itself. Nothing enables itself. Every action is audited.

Slack
Investigate CheckoutRetryError spikeStarted 09:41 · acme/checkout-api
Investigating
Demonstration — authored data
MC
Maya Chen#production-errors

CheckoutRetryError crossed the production threshold after the latest release. Can you find the cause before we roll back?

Demonstration on a fictional Acme repository — authored data and timings.

Call it fromSlackGitHubLinearSchedulesEchelar app

Code got faster.
Delivery didn't.

Every fix, report, and small change still sits in Slack or the backlog, waiting for the two people who know that subsystem.

A copilot suggests a change. The rest of the job is still yours:

  1. LOCALStill yours
    Run it

    Pull the branch. Reproduce what the AI claims.

  2. CIStill yours
    Test it

    Wait for the pipeline. Read the failures.

  3. PATCHStill yours
    Fix it

    Repair what broke, again.

  4. REVIEWStill yours
    Review it

    Read every line nobody fully trusts.

  5. MERGEStill yours
    Ship it

    Own the call, and the pager.

The suggestion was free. The shipping wasn't.AI can build it. Echelar helps your company build it right.

Echelar closes that loop.

One governed path, wherever the work starts. Every demonstration on this page follows it.

  1. 01Signal

    You call it.

    Mention it in Slack, assign it a Linear issue, open a pull request, or set a schedule.

  2. 02Sandbox

    Its agents do the work.

    In an isolated sandbox, grounded in your repository's architecture and conventions. Watch every step live.

  3. 03Evidence

    It shows its work.

    A pull request comes back with evidence: what changed, what was checked, what the checks said at that exact commit.

  4. 04Decision

    Your team decides.

    Merge, request changes, or close it. Echelar has no merge button. We didn't build one.

Work that starts without you.

Describe the recurring job in plain English. Echelar drafts the Automation: trigger, scope, checks, and where the result goes.

On a schedule

Sweep dependencies weekly. Re-sync docs nightly.

The same four steps, on a timer. When the job changes code, it comes back as a pull request.

On an event

Alert posted. Pull request opened. Tag pushed.

Automations fire from what already happens in your tools. No new dashboard, no new process.

Within bounds

Versioned. Audited. A person enables it.

Every fire leaves an audit record with a match trace and the exact prompt it sent. Any version can be rolled back.

Named agents. Named limits.

Not one chatbot: a team with roles, and every role has something it can never do. That column is the product.

AgentWhat it doesWhat it can never do
OrchestratorTakes your request, plans the work, and hands it to the specialists.Enable anything. Every Automation or Skill it drafts lands disabled.
Incident CommanderInvestigates an alert against the code, recent changes, and its own sandbox; posts a cause and a recommendation.Change code before a person says go.
Technical LeadPlans the technical approach and checks the Builder's work inside the Session, against how the repository already does things.Merge, enable, or widen the scope without a person.
BuilderImplements the change in an isolated sandbox, the way the repository already does things.Touch production or merge. The result is a pull request.
PR ReviewerReads each pull request against your repository and files findings by severity.Write code or merge. A fix is a separate Session a person starts.

Specialists work behind these — Explorer, Browser, and the Reflector that drafts a Skill when your team corrects the same thing twice. The product lists every agent and its version per repository.

We build Echelar with Echelar.This team works our own repositories every day, under the same limits.

It gets better on purpose. Never in secret.

When your team corrects the same thing twice, Echelar drafts the fix as a versioned Skill: disabled, evidence cited. When the code under a memory changes, the memory is demoted.

Repository learning recordAuthored · billing-api · three reviewed outcomes
First correction retained
Reviewed outcomesSame correction · 3 matches
  1. Session 184
    Migration guard missingReviewer finding
    repo@7f31c2PR #842
  2. Automation 071
    Migration guard missingReviewer finding
    repo@81a0bcPR #901
  3. Session 193
    Migration guard missingReviewer finding
    repo@8a44e1PR #918
Proposed improvementProposed as draft · disabled
skill / v3

Billing migration guard

Require a rollback path and verify schema compatibility before a billing migration returns for review.

Source
3 reviewed outcomes
Scope
Billing module
Revision
Skill v3
Priya enabled v3Human review · August 14
Future SessionsBilling work loads Skill v3 with the current repository context.enabled revision only

The questions worth asking first.

What can Echelar write to?

A branch in an isolated sandbox, and the pull request it opens from it. Merge, deploy, and enabling an Automation are actions that exist only for people; there is no capability for an agent to perform them.

What do we need to start?

The GitHub App on one repository, Slack or Linear connected, and your own model keys, stored encrypted. The first real alert or ticket is the test.

Does it work with our stack?

Anything that posts to an enrolled Slack channel can start an investigation, and the sandbox uses your repository's own environment definition. Echelar works from the alert, the code, and its history — it does not need access to your observability tools.

Where does our code run?

On Echelar's control plane, in an isolated sandbox per Session, scoped to your organization. Not inside your infrastructure. The security page has the exact answers.

How do we stop it mid-task?

Every Session is live: tool calls, changed files, checks. Reply, or close it. Code stays in the sandbox until it returns as a pull request.

What does it cost?

Echelar is launching now. Early teams work directly with the founder and agree pricing together.

Granting repository access is a serious question.

Here are serious answers: where code runs, who can enable what, and what we don't claim.

  • An isolated sandbox per session, scoped to your organization.
  • Your own model keys, stored encrypted. Per-organization GitHub tokens.
  • Per-organization roles and a full audit trail.
  • Data stays organization-scoped. Nothing is pooled across customers.
Read the security overview

Bring the repository you'd never trust to a demo.

Not the toy repo. The one held together by tribal knowledge. We'll point Echelar at it live. You watch what it reads, what it does, and where it stops to ask.