Skip to content

AI · Connect

Integrations

6 min read

Integrations are what the Console's Connect → Connectors page manages: remote services such as GitHub, GitLab, Jira, Canva, ElevenLabs, Datadog and Kubernetes, each with a declared sandbox tier, that your agents are meant to call (their calls don't work yet; see below). The Console calls them connectors; these docs call them integrations so they aren't confused with the engineering apps' connectors, which are functions you deploy into your own cloud for Terminal, Kubernetes and Data.

Status on September 25, 2026: OpenEng AI isn't live yet

The Console and Web Console interfaces load, but the OpenEng AI API routes and the Broker aren't deployed. Nothing can be saved or published, and no AI engine can register or connect. Pro, Enterprise and pay-as-you-go launch in Q4 2026 and can't be purchased yet. This page describes OpenEng AI as it works once those services are live (status).

Connect → Connectors — your integrations and the catalog; each is a remote https endpoint with a declared host-isolated sandbox tier, edited in a slide-over and picked per config in the builder.

What an integration is

An integration is a definition that points at a remote https:// endpoint. The engine on your machine only dials that endpoint; the integration's code never runs on your machine. Every definition also names a declared host-isolated sandbox tier, Tier 2 · jail (tier2Jail) or Tier 3 · micro-VM (tier3Microvm), for where the endpoint runs. The tier is a field the definition's author fills in: the engine checks the label, the https:// scheme and that the endpoint isn't a private address, but it can't verify how a remote host is isolated. The catalog's entries point at OpenEng-hosted endpoints under connectors.openeng.app/fn/, so the tool arguments your agents send them would go to OpenEng.

This rule, sandbox-only, is enforced three times: the Console's editor refuses a definition that breaks it, the API refuses it when you save, and the engine rejects it when it loads the config.

AspectIntegrationCellMCP server
Runsa remote, host-isolated endpointinside your own cloud accounton or near your machine, or any HTTP host
You deploy itno, the definition points at an endpointyes, into your cloudstdio servers run locally; HTTP servers anywhere
Transporthttps onlyhttps (plain http only to loopback)stdio or HTTP

The glossary sets connectors, cells and integrations side by side.

The catalog

The shipped integrations are published at connectors.openeng.app/integrations (machine-readable at https://connectors.openeng.app/integrations/catalog.json). As of 2026-09-25 the catalog has seven:

IntegrationCategoryToolsSign-in methodsTier
GitHubSource Control5: repositories, issues, pull requests, Actions runs and logsOAuth, API keyTier 2
GitLabSource Control6: projects, issues, merge requests, pipelines, jobs, job tracesOAuth, API keyTier 2
JiraProject Management5: issue search, issues, comments, sprints, boardsOAuth, API keyTier 2
CanvaDesign4: designs, a design, assets, export URLsOAuthTier 2
ElevenLabsMedia & Voice4: voices, models, text-to-speech, historyAPI keyTier 2
DatadogObservability3: metrics query, monitors, log searchAPI keyTier 3
KubernetesInfrastructure5: pods, pod logs, deployments, events, rollout statusAPI keyTier 2

The catalog tools are read-oriented (list, get, search, query), plus ElevenLabs' speech synthesis.

The Connectors page: Mine and Catalog

The page has two tabs:

  • Catalog — the shipped integrations, read-only. Use adopts one into your collection in a single click; Customize opens a copy in the editor first, under a fresh id you can edit and version.
  • Mine — your own integrations, adopted or written from scratch, listed with name, endpoint, sandbox tier, credential placeholder and tool count. Rows carry Versions, Edit and Delete (delete behind a confirmation; published config versions keep their frozen copy).

New opens an empty editor. Copies that the config builder pins to a config are listed separately under Pinned to configs, never mixed into your list.

Create or customize an integration

The editor is a slide-over with these fields:

FieldWhat it holds
NameRequired, e.g. GitHub
SlugA stable machine name; defaults to the id
Versione.g. 1.0.0
Endpoint URLRequired; https:// only
Sandbox tierTier 2 · jail or Tier 3 · micro-VM
CredentialA ${ENV_VAR} placeholder, never a value
Category, SummaryShown in the list
ToolsThe tool names the endpoint exposes; for information only, since the engine discovers tools itself
Auth methodsFor information, e.g. oauth, api_key
InfoLonger notes

Like every Console editor, it saves only when you press Save. There is no connection test: the Console stores configuration only, and the engine checks connectivity when it serves a config that uses the integration.

Use it in a config

A config uses the integrations you pick in the config builder's Connectors step (Configs); the step lists only your own integrations, so adopt a catalog entry first. Publish the config.

When the engine serves it, it admits the config's integrations that pass the sandbox-only rule and logs, for example, connectors registry — 2 admitted, 0 rejected (sandbox-only). A rejected definition never reaches your agents; the reason reads like connector '<id>' rejected: transport must be an https:// remote endpoint.

When an agent calls an integration's tool, the engine sends an https POST to the endpoint URL plus the tool name, with the arguments as JSON. It refuses endpoints that resolve to private or internal network addresses, and gives each call 30 seconds (10 seconds to connect). The call passes the engine's gateway like every other resource, so plugins on the connector gateway can gate or observe it.

Integration calls don't work yet

Two things limit integrations in the current release. The catalog entries point at endpoints under https://connectors.openeng.app/fn/, which return a web page instead of JSON (checked 2026-09-25). And the engine doesn't send the credential you store with a definition yet; calls carry only the tool arguments. Once the OpenEng AI services are live (status), you can author, adopt and publish integrations on every plan; calling them needs an endpoint that answers.

Plan & tier notes

The Connectors page is open on every plan. Running a config needs the openeng engine, which registers only with an active Pro plan; Pro launches in Q4 2026 and can't be purchased yet (Plans & billing).

Troubleshooting

"The sandbox tier must be tier2Jail or tier3Microvm." Pick one of the two host-isolated tiers; lower tiers would run on your machine and are never allowed.

The editor refuses the URL. It must parse and start with https://.

The builder's Connectors step is empty. It lists only your own integrations. Adopt one on the Catalog tab (or create one), then return to the step.

The engine log shows rejected (sandbox-only). A definition in the config breaks the rule (not https, or a tier below Tier 2). Fix it on the Connectors page and publish a new version.

A tool call fails with connector … decode: …. The catalog endpoints return a web page instead of JSON, so the engine can't decode the answer. See the warning above. A connector … → HTTP 404 (or another status) means the endpoint answered with an error, for example because it doesn't serve that tool.