We’re building Ever Works in public. Some things may be incomplete, missing, or broken while we continue improving the platform.We’re building Ever Works in public — expect a few rough edges.
Operate · Chat does everything
The dashboard has hundreds of buttons. The chat rail has the same operations as tools — generated from the platform’s own API, one tool per operation — so creating a Mission, assigning a Task to an Agent, running an Agent now or setting a budget is something you can say instead of something you have to find. Every call runs as you, with your permissions, and the API enforces ownership exactly as it does for the UI. Anything destructive stops on a Confirm card first. Results come back as charts, tables, boards and detail panels on a side canvas, not as paragraphs describing them.
One chat tool per platform operation, generated from the same API the dashboard calls — so chat coverage is a data edit rather than a new feature build every time the platform grows.
Over three hundred tools are generated straight from the platform’s operation registry, alongside around sixty hand-written ones for the paths that deserve bespoke handling. Create a Mission, assign a Task to an Agent, run an Agent now, set a budget, deploy a Work, rotate a webhook secret — each is one sentence, and each runs as you.
Delete, revoke, disconnect, cancel and rotate-secret operations are flagged in the registry: instead of performing the change, the tool returns a Confirm / Cancel card, and the mutation runs only after you click Confirm. Fifty-five operations are gated this way today.
Bulk endpoints are never registered, and a runtime guard rejects any call that tries to smuggle an array of ids through a request body. There is no "delete all my Works" — by design, not by omission.
Charts, tables, stat tiles, detail panels and kanban boards render into a side canvas, plus twenty-one bespoke components — timelines, gauges, funnels, heatmaps, calendars, code and Markdown blocks, and typed agent questions you answer in place. Every result keeps a chip that reopens its artifact later.
Dozens of named reports ship built in — tasks by status, missions by status, spend over time for a Work, spend by plugin, activity per day, webhook deliveries by status. When none of them fits, the report builder groups any list of your entities by any field and charts it.
When your deployment has a transcription provider configured, a mic appears in the composer; dictated text is appended to what you already typed, never sent on its own. Drop files onto the composer and they attach as reference data — and are also handed to Memory in the background, so the org keeps them after the conversation scrolls away.
The rail is one surface, but a conversation always knows what it is about — a Work, a Task, or the account as a whole — and the same chat is reachable from outside the dashboard.
Opened from a Work, the conversation carries that Work’s id and the page you are on, so "what changed here" resolves to this Work. Answers drawn from the Work’s Knowledge Base render a citation footer under the message, linking to the documents they leaned on.
Every Task has its own message thread. Write @agent-slug to bring one of your Agents into it — the mention is resolved server-side and dispatches that Agent to reply on the thread — and [[kb]] tokens to point at knowledge-base documents. Mentions that resolve to nothing are dropped rather than guessed at.
Conversations are stored and titled, and listed behind the History button; pick one up days later, or start a clean thread with New chat. The rail itself remembers whether it was open, how wide it was, and whether you had it expanded.
Mention @works in a connected Slack channel, or run the /works slash command: you get an instant private acknowledgement and the answer posts back into the channel, every delivery signature-verified. For your own clients, POST /api/v1/chat/completions is OpenAI-compatible, streams, and accepts x-work-id and x-provider-override headers.
Four gestures — and the rail comes back the way you left it every time.
Click the bot toggle on the edge of the chat panel in the dashboard. Drag its border to resize it, or expand it to full width when a canvas artifact deserves the room.
The provider sits in the chat toolbar — it is the thread’s identity. The model sits in the composer, because "this one needs the big model" is a per-message call. Leave it on provider default to inherit whatever your admin configured for the tenant.
"Create a mission for the cats directory." "Show me spend by plugin for this Work." "Pause the researcher agent." Type it, dictate it if your deployment has a transcription provider, or attach a PDF and ask about that instead.
A destructive call stops on a Confirm / Cancel card until you decide. Everything else answers in the thread with a chip that opens the artifact — a chart, a board, a detail panel — in the side canvas.
Chat coverage is a registry, and the registry grows wave by wave. Roughly four hundred tools ship today against a mapped inventory of every operation the platform exposes, so a handful of newer endpoints still have no tool of their own. Two more things worth knowing: each turn offers a bounded, relevant slice of the tool set — the always-on core plus the domains your message and current page match — so a domain that was not obvious in one message comes straight back when you name it; and when no tool exists for what you asked, the chat can still take you to the Works, plugin or settings screen that has the button — that fallback is a fixed set of destinations, not a route to every screen in the product.
Run many agents at once from one cockpit — live status, current activity, tokens and cost — and steer, interrupt, or resume any of them.
Learn more →Drive the platform from Slack, get pull requests reviewed against your knowledge base, and turn meetings into memory.
Learn more →Drop in PDFs, Office files, web pages, Notion pages and recordings — and get typed, Git-backed knowledge with hybrid search and answers that cite their sources.
Learn more →Everything the dashboard does is an authenticated REST call — documented with OpenAPI, callable with an API key, and observable through signed webhooks.
Learn more →Manage Works from any MCP-compatible assistant — 127 API tools plus Knowledge Base tools, authenticated with an API key, over stdio or HTTP.
Learn more →Agent questions, approval requests, escalations and notices land in one Inbox — and reach you in-app or on your own chat channels, tuned per event.
Learn more →Describe what you want to build. Ever Works researches it, ships it, and keeps it improving — content and code, owned in your own Git.