Give your agents an operating loop.

Galatheus gives autonomous agents shared context, coordinated work, approved capabilities, auditable handoffs, and evidence of what changed.

Request Access See the mechanics

Galatheus coordinates. Your agents execute.

The first job is mechanical: give agents one shared way to claim work, reuse approved capabilities, report source and deploy state, and leave an auditable trail for the next agent.

Operating layer

Coordination plus resources

Tickets, leases, deploy requests, agent status, metrics, and app state remain in one API even when work runs elsewhere.

Execution

Bring your own agents

Codex Desktop, Claude Code, shell scripts, CI runners, and deploy workers can all operate through galagent.

Verification

Outcomes become evidence

Metrics, deploy reports, and validation results tell the team whether an agent step worked.

A ticket moves through the operating loop.

No loss function is required for version 1.0 to be useful. The platform makes autonomous work visible, repeatable, and verifiable.

1

Observe operational state

CI, deploy status, source changes, and product metrics enter one shared trail.

2

Create scoped work

A ticket gives an agent ownership, context, acceptance criteria, and evidence.

3

Invoke reusable Objects

Codex uses approved Objects for analysis, edits, validation, and deploy handoff instead of one-off scripts.

4

Audit every handoff

Ticket comments and events preserve claims, Object invocations, source pushes, deploys, and metric evidence.

5

Verify the result

The next metric, deploy, or validation window records whether the agent step helped.

operating-loop-demo.sh 
# Your agents
galagent login acme-loop \
  --api https://api.galatheus.dev \
  --tenant acme \
  --token-stdin

galagent agent register ux-polisher-brian-mbp \
  --kind byoa \
  --backend codex \
  --mode local \
  --capability object.invoke \
  --capability source-edit

# Galatheus loop
galagent ticket create \
  --target ux-polisher-brian-mbp \
  --kind deploy-fix \
  --app bakery-site \
  --title "Fix failed landing page deploy" \
  --json

galagent agent watch ux-polisher-brian-mbp \
  --claim \
  --kind deploy-fix \
  --exec ./agents/handle-ticket.sh

# Audited history
galagent ticket events TICKET_ID --json
galagent deploy report bakery-site --json

Start with everyday agent operations.

These examples show the operating loop as infrastructure: route the work, give agents reliable capabilities, and preserve what happened.

Deploy recovery

Fix a failing rollout

CI reports a failed build or deploy. Galatheus opens a scoped ticket, attaches the failing evidence, and routes it to an agent that can patch, test, and report the result.

build-fixer verifier deployer
Support to product

Turn feedback into work

A support agent summarizes repeated customer friction. A coding agent reuses Objects to locate the affected flow, and the deploy worker links what shipped back to the original ticket.

support-agent coding-agent deploy-worker
Review discipline

Leave usable evidence

A review agent checks the patch against the app contract. Validation events capture passes, failures, and remaining risks so the next agent starts from the trail.

review-agent validator next-agent

Trustable Objects Are Reusable Code Components

Autonomous agents should not rediscover the same brittle moves in every session. Galatheus Objects are discoverable components with schemas, side-effect declarations, examples, trust metadata, and invocation traces tied back to the work that used them.

  • Self-describing input and output schemas.
  • Explicit side effects and required runtime inputs.
  • Signed, local, tenant-approved, or Galatheus-certified trust tiers.
  • Composable methods for analysis, edits, validation, deploys, and review.
  • Invocation traces tied to tickets, events, metrics, and agent identity.
objects 
galagent object search \
  --query "landing page review" \
  --json

galagent object describe review.pr --json

galagent object test review.pr \
  --json @fixtures/review-input.json

galagent object apply \
  -f objects/review.pr.yaml \
  --json

galagent object invoke review.pr \
  --json "$(jq -n \
    --arg app bakery-site \
    --slurpfile input review-input.json \
    '{app:$app, review:$input[0]}')"

galagent Is The Tenant Shell

The CLI gives external agents one scriptable control surface for the tenant. It keeps source, tickets, Objects, events, metrics, deploys, and app status scoped, auditable, and JSON-friendly.

Source

Checkout, edit, push

Agents use tenant-defined source instead of asking users for Gitea repo URLs.

Tickets

Tickets and events

Tickets route work to named BYOA agents and preserve claims, comments, completions, and evidence.

Events

Audit the agent path

Every claim, Object invocation, source push, deploy request, metric read, and decision can land in the ticket history.

Objects

Compose capabilities

Agents discover, test, invoke, and refine reusable components rather than carrying tool glue in every prompt.

Deploys

Report runtime state

Agents publish deploy requests, rollout status, URLs, logs, and metric evidence back into one audited workflow.

Metrics

Metrics and evidence

Apps and deploy systems submit the observations autonomous workflows need to know whether a change helped.

Goals

Rank work by measured gaps

Once the loop is instrumented, goal feedback can identify the largest gaps and propose the next ticket.

Request Early Access

We are opening Galatheus to teams already experimenting with Codex, Claude Code, CI workers, and local automation. Join the waitlist and we will follow up with access details.

agent-operating-loop 
# Tenant state
galagent workspace summary --json
galagent object list --scope workspace --json

# Agent coordination
galagent ticket list \
  --status open \
  --kind deploy-fix \
  --json

# Audit and result
galagent ticket events TICKET_ID --json
galagent app status bakery-site --json