0.3.14

• MCP servers are named after their app. A server from the MCP Registry with no title (Stripe's com.stripe/mcp, for one) showed as "mcp" on the plan review and was connected as the ability mcp-mcp. It's now named after its app or publisher: Stripe, mcp-stripe, STRIPE_MCP_KEY. An app already connected as mcp-mcp keeps that name; disconnect it on the Apps page and connect it again to rename it.
• Clearer note about approvals on the plan review: when a workflow needs your approval or an answer, it asks in Chat if you're running it there, and otherwise sends the request to the app on the agent's Messaging tab.

0.3.13

License change. From this version, LoopEngine is under the Functional Source License, Version 1.1, MIT Future License (FSL-1.1-MIT):

• You can use, change and self-host it for anything, including running your own business on it, except offering it to others as a competing commercial product or service, such as a hosted LoopEngine.
• Each version becomes MIT-licensed two years after its release.
• Versions up to and including 0.3.12 were released under MIT and stay under MIT.

Third-party software included in LoopEngine keeps its own license: see THIRD_PARTY_NOTICES.md. No code changes in this release.

0.3.12

• Every workflow example on the New workflow page. "Start from an example" lists 116 examples across 21 areas, as on loopengine.co: the seven ready-made ones open their plan straight away, and the rest fill in the description for you to change and plan.
• Each card shows the apps it works with and what it asks you before; search, filter by area, and page through them six at a time.

0.3.11

• Workflow examples instead of templates. The New workflow page (one sidebar entry, replacing "Start from a template" and "Describe a workflow") has ready-made examples: opening one goes straight to the plan review, with no model call, and you can create it as a new agent or add it to an agent you have. Each example's schedule runs its workflow, so runs get history, Run now and webhooks.
• The review screen shows a new agent's instructions, editable, and the apps, skills and setup notes a plan comes with.
• Plans can bring skills and settings, such as a policy you fill in; an agent that already has one keeps its own.
• The planner writes better instructions for a new agent: its role, things to fill in, and rules that always hold, never the steps.
• /templates now redirects to the examples. A project server from an older template keeps working.

0.3.10

• Official abilities only. Of the public abilities on npm, only LoopEngine's own (the @loopengine-co scope) are listed and can be installed. Older lp-* packages are refused; their official versions are under the scope. Private abilities are unchanged. LOOPENGINE_ALLOW_UNOFFICIAL_ABILITIES=1 turns the check off.
• Templates on official abilities and apps. Policies skills you fill in instead of docs search, the Firecrawl app instead of web search, @loopengine-co/hubspot, gmail and shopify, and results sent through the Messaging tab instead of a Slack posting tool. Incident triage is now a workflow to start from your alerting tool's webhook.
• Your own MCP servers. The Apps page adds a company's own server by its address, with an optional key and header. It connects like any app and only your workspace sees it.
• "Send each result" now reads "Message me the result when it finishes".

0.3.9

Messaging

• Each app's Approvers (Slack, Telegram) are now inside that app's section on the Messaging tab, under its connection, not at the bottom of the page.
• Telegram's Check connection says when its last delivery error happened and how many updates are waiting. Telegram keeps the last error until a newer one, so an old error no longer looks current.
• A job refused by direct message points to Telegram → Approvers.

Console

• Tabs: Overview, Workflows, Apps, Chat, Messaging, Schedules, Permissions, Environment, Advanced. Workflows and Schedules show a count from the moment an agent opens; Permissions no longer does.
• The Overview no longer shows the Hooks section or skillsDirs, and Advanced → Skills no longer shows Skills dirs.

OpenAI

• A reasoning model (GPT-5 on, the o-series) with no reasoning effort set now uses low, with the 16k output limit. Unset, it reasoned at OpenAI's default inside a 4,096-token limit and could use it up before calling a tool. Older models (gpt-4o, gpt-4.1) are unchanged.

Workflows and templates

• Permissions are set after abilities and apps are installed: no rule is added twice, a stricter one (ask where an ability allows) is placed so it decides, and a decision the agent already has is never loosened. Rules are named after their tool.

0.3.8

Workflows: clearer names, one editor.

• The workflow editor has a Description field: the one line the agent uses to recognise the job when you ask in Chat. It also shows the workflow's ID read-only under Title (used by the API, schedules and webhooks; it never changes).
• A workflow whose title is just its ID now reads as words, and the workflow planner turns a name written as an ID into words.
• In Advanced → Skills, a workflow is labelled Workflow, and its Edit opens the workflow editor on the Workflows tab. Plain skills keep the skill editor.
• The Schedules tab shows a workflow's description under its title, and the title links to its card.

0.3.7

Workflow cards, continued.

• How it starts is now Ways to run it.
• Each schedule on a card links to its row in the Schedules tab, scrolled to and highlighted (also as ?tab=schedules#schedule-<id>). Its name in brackets is gone, it says results go to your messaging app when it sends them, and Paused is highlighted.
• The API section always shows a body worth copying: the workflow's params and "wait": true, with a line on params, wait, notify and callback_url.
• The schedule form's hint speaks of the workflow's instructions, not its start message.

0.3.6

Clearer workflow cards.

• How it starts is its own panel with four sections, each with a ? that explains it: Schedule (with Add a schedule), Webhook, Run now (with the values a run is given) and API (the exact call your app makes). Recent runs, Edit and Remove workflow follow; Instructions are open by default.
• Forget is now Remove workflow, and it also deletes the workflow's webhooks. Its instructions stay as a skill the agent can use in Chat.
• A workflow with nothing starting it says No schedule or webhook yet.
• Clicking an agent in the sidebar opens its Overview (from Chat you stay in Chat), and Chat is now the last tab.

0.3.5

Workflows, simpler.

• One Instructions text. A workflow's start message and steps become one text you edit as one (the skill body, marked workflow: true). A run starts with a line naming the workflow, then the instructions. Older workflows show their start message and steps together and are converted the next time they're saved.
• How it starts, for real. The "When it runs" label is gone: each workflow card lists its actual schedules (in words, in the workspace's time zone, with Pause/Resume/Remove), its webhooks, and Run now or the API, with Add a schedule and Webhooks.
• A Workflows tab next to Overview. Overview gets a Workflows tile (count, next run, waiting for a decision); Tools and Skills share one tile.