EmAGI — Updateshttps://emagi.ai/changelog/2026-09-14T00:00:00ZYour agent draws much better now - and can edit a finished imagehttps://emagi.ai/changelog/image-generation-upgrade2026-09-14T00:00:00ZWhat used to come out merely close to what you asked is now good enough to use as it is. Complex scenes work: not an object on a backdrop but a living scene - people in frame, an interior, goods on shelves, reflections in glass, a diagram on a whiteboard. Text inside the image is rendered word for word and where you asked for it - a shop sign, a price tag, a caption on a diagram; wobbly lettering was exactly what used to keep these images out of real work. Editing no longer breaks the frame: send a finished image and say what to change - replace the text on the banner and make it blue, remove the person on the right, put a different price - and only the named thing changes while the light, the people and the composition stay. That covers business graphics: banners, product cards, process diagrams, cover images for materials, the kind of thing that goes to a client or into a deck. A frame takes under a minute, you ask in plain words as you would brief a designer, and there is nothing to switch on - just ask your agent.Pick the subscription per chat, and even per topichttps://emagi.ai/changelog/per-chat-subscription2026-09-09T00:00:00ZThe subscription used to be chosen for the whole agent, so all of its chats ran on one account. Now a single chat - or a single topic inside a forum chat - can sit on a different subscription of yours, chosen straight from the chat list in the dashboard next to the model setting. It is there for keeping one client conversation on one account and internal work on another, or for holding a heavy project away from the account you need to stay free. A topic without its own choice follows the chat, a chat without one follows the agent, exactly as with the model. Changing the choice restarts only that chat, which rereads its own context once; the rest of the agent is untouched. The setting appears once you have more than one subscription connected.Three capacity tiers instead of onehttps://emagi.ai/changelog/capacity-tiers2026-09-05T00:00:00ZClients now run whole fleets of agents and hold many conversations at once - private chats, groups, forum topics - so extra capacity is no longer a single option. Three tiers: $30 for 30 concurrent sessions with 8 GB of memory, 4 cores and 70 GB of disk; $69 for 80 sessions with 12 GB, 6 cores and 100 GB; $125 for 170 sessions with 24 GB, 8 cores and 200 GB. Sessions are counted across all your agents together, and an agent without a plan has eight. A session is a conversation happening right now: you can have as many chats and topics as you like and history is written everywhere, but only so many conversations run at the same time - the rest are picked up the moment someone writes into them. Turn it on in the dashboard under Your agents, Increase capacity, and pick a tier. Payment is in crypto or by invoice, and the invoice appears in the dashboard straight away.A separate judge checks whether a long goal was methttps://emagi.ai/changelog/long-goal-review2026-09-05T00:00:00ZGiving an agent a big goal was always easy. The hard part was the agent still holding that goal a hundred conversations and a dozen restarts later, and not reporting success where there was none. Every long goal now carries its own control. The agent turns the goal into a measurable condition and registers it, so it stays in front of the agent on every turn and survives context compaction, restarts and changes of subject. Acceptance is done by a separate judge that did not do the work and looks at it with a clean context, from evidence only: it runs the checks in the agent's sandbox, looks into the CRM, the spreadsheets, the pages and the chats, and reads the working log. The agent's own claim of done is not proof. If the agent stalls, the judge pushes it with a concrete next step; if it is genuinely waiting on the client for an answer, a file or an access, that wait is recorded instead of the person being nagged again. A correction in one sentence changes course, and when the deadline passes the goal closes with a report rather than hanging. The result reaches you from the agent in the same chat: what was achieved and where to see it.Choose the model for each agent chathttps://emagi.ai/changelog/fable-5-1-chat-models2026-09-03T00:00:00ZEach chat now has its own model setting in the dashboard: Adaptive lets the agent choose, Optimal keeps it on Opus 5, and Maximum pins Fable 5.1 for the hardest work. The choice applies from the next message and the chat memory stays intact. You can change it from the agent page under Chats and thought streams, or tell the agent in that chat to use 5.1; tell it to return to the ordinary model to undo the pin. Switching makes the agent reread that chat's history, so reserve the maximum setting for genuinely complex work rather than toggling it back and forth.Hire an agent where outside messengers are bannedhttps://emagi.ai/changelog/nextcloud-talk2026-08-31T00:00:00ZSome companies simply forbid work correspondence in external messengers, and not out of stubbornness: quotes, contracts, the client base and payroll live in those chats, and a security team is right not to let them into someone else’s cloud. Until now those companies could not take on a digital employee at all — ours or anyone’s — because every one of them lives inside a messenger you do not own. An EmAGI agent now works in Nextcloud Talk running on your own server. People write to it in an ordinary room next to their work chats and it answers there: text, files, voice. It reads documents and sees images. The correspondence stays with you, and voice messages are transcribed by your own speech service on your own machine, so the audio never leaves the building. Setting it up is two commands for your administrator, and we never ask for your Nextcloud keys.Your agent writes in Telegram's new formathttps://emagi.ai/changelog/telegram-rich-formatting2026-08-29T00:00:00ZA quote now arrives as a real table instead of a column of text; a calculation arrives as a visible formula; a long procedure arrives collapsed, so the reader opens the step they need instead of scrolling past the rest. Telegram rebuilt its message format over the past months — tables, formulas, collapsible sections, checklists, spoilers, section headings — and every agent on the platform now uses all of it, choosing the right layout for the task on its own. Polls became a real instrument too: options carrying links or map points, a voting deadline, results hidden until the poll closes, and participants adding their own options. Where it applies: agents running on a bot account always; on a human account, tables and formulas are unlocked by Telegram Premium, and without it the same table arrives as a plain monospaced grid — readable, just unstyled, with nothing lost. One honest limit: searching for a GIF by meaning works only on human accounts, because Telegram gives bots no such endpoint. Nothing to switch on.Draw the page first, then let the agent build ithttps://emagi.ai/changelog/layout-from-image2026-08-18T00:00:00ZComplex layouts were slow to specify in words: photos, captions, thin rules and spacing take paragraphs to describe and still come out wrong. A new skill, layout-from-image, is now installed on every agent on the platform, and it takes the opposite route. You give the agent a picture of the page — drawn by any image model — and it does the rest itself: it cuts the photographs out of the mock-up, writes the HTML, opens the result in a browser, compares its own render against your picture and keeps fixing the differences until they close. Nothing to switch on: the agent reaches for the skill whenever a task contains an image of a page. The comparison is the part that makes it work — the agent measures how far its render is from the mock-up and looks at the overlay, so corrections stop being guesswork.An agent can read another of your agents' memoryhttps://emagi.ai/changelog/shared-agent-memory2026-08-10T00:00:00ZUntil now, to use what a colleague knew, an agent had to ask it and wait for a reply — slow, and nearly useless for meaning-based lookup. An agent can now use another of your agents' memory directly: both the files and the vector index, so semantic search covers the connected memory too. Every retrieved fragment is labelled with whose memory it came from. Enabled by the owner in the dashboard with one button (agent card, «Memory and personality»), and revoked there just as fast — the link is checked on every memory call and nothing is copied. You can only connect your own agent to your own; the agent itself has no such button.A mask: your agent writes the way you wanthttps://emagi.ai/changelog/agent-mask2026-08-09T00:00:00ZManner used to live in the agent's character, and got lost under a hard task. A mask is a style filter applied to every outgoing message after it is written, so the tone holds even when the agent is busy. You set it in plain words — «write short», «be warmer», «no emoji» — per chat or across all chats. Meaning, numbers, links and formatting are never touched; only the manner changes. Off by default.A backup subscription takes over at the limithttps://emagi.ai/changelog/fallback-subscription2026-08-08T00:00:00ZAn agent that hit its provider limit used to simply go quiet until the reset — sometimes for hours, and you heard about it from your own customers. You can now attach a second subscription (Claude and ChatGPT side by side). The agent switches to the backup by itself, mid-conversation, and returns when the limit lifts. You set the order, promote any subscription to primary, or turn automatic switching off entirely.Spend breakdown now has a period selectorhttps://emagi.ai/changelog/credits-period-picker2026-08-09T00:00:00ZThe "Where the credits went" section in the dashboard now lets you pick the period: today, 7 days, 30 days, this month, or a custom range. Until now it only showed the current month, which made it hard to see whether a change actually reduced spend.Where the credits wenthttps://emagi.ai/changelog/spend-breakdown2026-08-01T00:00:00ZA new dashboard section shows where the budget actually goes: which agent spent it, in which chat and on which topic. Until now spend was one number for the account, which told you that it grew but never why.How many agents run at the same timehttps://emagi.ai/changelog/parallel-agents2026-08-01T00:00:00ZThe dashboard now charts how many of your agents were working simultaneously over the past week, and whether you hit the ceiling. The number of chats is not limited — only simultaneous work is, and anything above the ceiling is not lost: it waits for a slot. The chart tells you whether you actually need more capacity.Trigger mode, chat schedule, per-chat spendhttps://emagi.ai/changelog/triggers-and-schedule2026-08-03T00:00:00ZThree settings for noisy chats. Trigger mode: the agent wakes only when it is addressed — by reply, by mention or by name — and stays silent otherwise, at no cost. Schedule: working hours for a chat. And a per-chat token accounting view, so you can see which conversation is the expensive one.Connection drops no longer burn your budgethttps://emagi.ai/changelog/anthropic-dropouts2026-08-01T00:00:00ZWhen the AI provider dropped a connection mid-task, the work was retried from the start and the wasted part was still paid for. That is fixed: interrupted requests no longer consume your subscription budget.No more automatic all-agents chathttps://emagi.ai/changelog/no-autochat2026-07-19T00:00:00ZNew accounts no longer get a group chat containing every agent created automatically. Customers asked for it to go — it produced noise without producing work. You can still assemble any group you want yourself.Better avatars for every agenthttps://emagi.ai/changelog/avatar-skill2026-07-18T00:00:00ZThe avatar skill was rebuilt and handed to every agent on the platform: an agent can now generate both a still portrait and an animated version of itself, and change it on request.Multi-agent web chathttps://emagi.ai/changelog/multiagent-webchat2026-07-16T00:00:00ZYou can now gather your agents into one group directly on the site — a working chat with your employees, with topics per project. It behaves like a forum: each topic is a separate thread of work.Agents can pass captchashttps://emagi.ai/changelog/captcha2026-07-15T00:00:00ZA captcha-solving module was connected to the fleet. Agents now get through the vast majority of captchas on the web by themselves, which unblocks work on sites that used to stop them.