ai

Added

Notey can answer comments on your own Bluesky and Mastodon posts. It is off by default. Each platform has its own mode (off, suggest or auto, and auto asks you to confirm), plus a rules editor, a test box that never sends, a list of recent replies with delete and never reply again, and a kill switch in the Inbox and in Settings. The daily digest says how many replies went out and how many are waiting for you.

Added

A Library panel ranks your X posts (archive imports, captures and posts NoteSurge published) by engagement and turns the one you pick into an Instagram carousel with a caption, saved as a draft. Nothing is posted automatically. Archived tweets can now become carousels too, and Notey and MCP clients can do the same.

Added

Break this post down, on a Library card, shows a post’s format, hook, beats, payoff and skeleton, with every quote checked against the post, and maps it onto an idea of your own. Anything it cannot find in the post is labelled not found, and it never invents numbers. The Library also has a chat that answers from your best-scoring pieces and quotes the lines it relied on, each one checked against the piece it names.

Added

Eight ready-made studios install in one click: X Post, Substack Newsletter, Weekly Content Planner, Topic Research, Script, Personal Brand Strategist, Thread and Repurpose. Routines gains four presets: a morning content brief, five post ideas in your niche, a weekly analytics recap and evergreen resurfacing.

Added

A new standing task searches the web for rising topics in your niche on a schedule, records what it finds, and queues up to three draft angles for your review. It never publishes anything.

Changed

The canvas now follows light and dark mode, including its zoom controls and edge labels. Each node is a fixed-width card with its kind, name, details and run status, so long labels no longer overlap, and nodes nothing links to are marked not linked.

Added

NoteSurge now learns what works for your account as a whole (length, media, links, topics and how far apart you post) from your own measured posts, once a platform has at least eight. Each recommendation names the posts it rests on, and a pattern you dismiss stays dismissed until the numbers behind it change.

Added

Notey can check an image before it goes out: what it shows, any text in it quoted exactly, and anything that looks wrong. It can also open a Library piece or a draft for you and read back the margin notes you leave on it.

Added

The workflow canvas’s Add node panel can create a new trigger or skill and place it on the canvas without leaving it. The Create a skill panel also gains a Save to my skills button.

Changed

Notey now summarises long conversations on Discord, Slack, WhatsApp and iMessage too, instead of only reading the last 20 messages. On Telegram, a reply that failed to send is delivered again later, marked as delayed.

Changed

The right rail, the AI tab and a new phone sheet (behind a floating Notey button) now share one conversation. Your message shows at once, the steps Notey took are saved with each reply, and replies can hold tables and drawn diagrams. An empty chat offers starter prompts, and replies have Copy, your messages have Reuse, and a failed turn has Try again. The chat only scrolls itself while you are at the bottom.

Changed

Analytics now has Overview, Growth, Engagement, Links, Ask your data, Report and Ideas views, each linkable, with the performance report called out at the top of Overview. Ideas explains why a post did well against your own typical post and builds an idea matrix of new angles to try.

Changed

The card menu is grouped into Write & repurpose, Media, Share & publish and Manage, with Compose and Ask Notey as buttons on the card. A piece with mermaid or excalidraw diagrams gets Draw the diagrams, which shows them as images without changing the piece. Viral rewrite drafts a new version shaped by the patterns in your own best posts.

Added

A Feature status section shows which optional features are available, and features that are not set up are turned off with an explanation instead of failing. Agent run control shows whether Notey is working and lets you stop it, with a confirmation.

Added

A skill can carry its own schedule. Installing it also sets up a standing task that produces drafts for your review, never published posts.

Fixed

Tighten and quick edit no longer replace a long draft with a cut-off reply. The reply budget grows with the draft, and a reply that still comes back cut short leaves your text as it was.

Changed

Notey’s answers now stream in as they are written, in the app and in Telegram. Long conversations are summarised instead of cut off, and a model that keeps failing is skipped for a while rather than retried every time. You can tell Notey to forget something it remembered, and scheduling a note that is nearly the same as one already queued asks first.

Added

Voice notes sent to Notey on Telegram are transcribed and answered like text, with spoken replies as an option (off by default). A new Spend card in the AI tab shows this month’s usage and lets you cap what runs on your own key.

Added

Settings can now pair a Telegram, Discord, Slack, WhatsApp or iMessage chat with Notey using a one-time code, list the paired chats and unpair them. The daily digest also reports anything that has stopped running or is stuck, on every channel you get it on.

Added

AI apps that support remote MCP can now connect to NoteSurge by signing in and approving access, with no token to copy. Tokens can be limited to analytics, one area or one platform, can expire, and Settings lists the apps you have authorized with a Disconnect button. MCP clients also get NoteSurge resources and prompts, and every API capability is now available over MCP.

Added

Two new standing tasks save a Library piece on a schedule: a weekly review of how your posts did, and a brief on what accounts in your niche are posting. Neither publishes anything. A starter pack of skills is available to install.

Added

Export PNG sits beside Export PDF on Library cards and newsletter drafts and saves branded images, one per page for long pieces. Make a podcast version turns a post, thread or newsletter into a one or two voice episode.

Added

When a video has no captions, Transcribe can fetch its audio and transcribe that instead, and tells you why if it cannot (private video, too long, or a worker error).

Changed

When you publish two versions of the same idea on Threads, the likes and reposts they get are now compared the same way X and Bluesky results already were, and the winning openings show up as a learned pattern in the AI tab. It still needs three measured head-to-heads before it says anything.

Added

Your own submissions now appear in the Skills tab with their status and any note from the reviewer.

Added

When you add a piece of 280 characters or more, Notey can write a short margin note on it in the background. It is limited to a few pieces per hour and ten notes a day, only runs on your own AI key, and respects the stop switch. Turn it off under AI, Agent behavior.

Added

Make an audio version in a Library card menu narrates a piece to an mp3, and Clean up the audio evens out the volume, trims silence and can add an intro and outro. Narration uses your own OpenAI key, or a text-to-speech service you point NoteSurge at.

Changed

Submitted skills go live straight away, up to five a day per person. Every listing is labelled as community submitted and unreviewed, installing asks you to confirm, and reported skills can be taken down.

Added

Members can comment on a draft inside NoteSurge and @mention each other (the mention goes to your team channel). A thread can be resolved and reopened, each draft has its own history, and you are warned when someone else has the draft open or saved since you did, with a choice to overwrite. In the Compose AI panel, Continue, Feedback and Tighten work from your voice sample and standing instructions.

Added

Pick your own From and To dates in Analytics. Review links unfurl with a card. Workspace owners can stop and resume agent runs from the AI tab.

Fixed

Both sub-tabs had been showing as bare icons since early September.

Fixed

loadQuoteCardPreferenceCounts and loadCarouselPreferenceCounts shared a context_entries row keyed only on kind + account_id, so whichever feature wrote last clobbered the other. Added a metadata JSONB containment filter to all 4 SELECT paths and changed the write path to a metadata merge (#766, PR #770).

Fixed

Cancel a Notey chat or AI post-generation request and req.signal now threads through the chat route and provider tool loops, stopping server-side generation instead of letting it run to completion (afk-519).

Changed

The Ledger (morning-ritual publishing timeline) moved from a top-level sidebar section into the AI tab as a peer sub-tab alongside Context, Agent, Models, MCP, and Knowledge — the app shell returns to a clean six-section sidebar (Compose, Queue, Library, Analytics, AI, Settings). Legacy ?section=ledger URLs forward to the AI tab.

Added

Creators can publish a skill from their workspace to a curated marketplace directory, where other users can browse approved entries and install them with one click via the same SKILL.md import path. Submissions start as pending and require admin approval before appearing in the public directory. Free directory in v1 — no payments, paid listings may come later.

Added

Asset generation stage 2: a rendered quote card is no longer a one-shot. The AI writes a draft subhead + body, then the card shows as a real at-a-glance image on its Library piece — click it (or "Review & edit quote card" in the ⋯ menu) to open an edit modal where you fix a weak hook or clunky line and Save. Saving re-renders the card through the deterministic satori template (same brand geometry, zero AI spend on the edit) and updates the stored marker + asset audit trail in one transaction. The original card is never touched on render failure, and empty/oversized/no-change edits are refused honestly instead of double-writing.

Added

The learning loop that used to live only in the repurposing engine now spans the whole product: in-app producers (compose, queue actions, standing rules) feed a unified event stream, the dashboard shows a legend of what is being measured, and future intelligence features start from real approved behavior instead of guesses.

Added

The AI co-writer that used to be Telegram-only now has a home in the app: a sidebar chat on every page (right rail). Same Notey, same memory of your creator profile, same standing rules — now one click away from the queue.

Added

Accounts run by agents land on the Ledger: a timeline of today's repurposing ritual — what the agent produced, what needs your approval, and the day's publishing plan — instead of a blank compose screen.

Added

NoteSurge could already expose an MCP server for agents like Claude Desktop to consume. Now the flip side: Notey itself can connect to any external MCP server, discover what tools it offers, and call one. The stateless client handles Streamable-HTTP JSON-RPC 2.0 (bare JSON or SSE framing), validates URLs before any network hit, and surfaces server-side tool failures honestly — never collapsing errors into fake success. Both capabilities are registry-tied: mcp.tools is a read-tier discovery, mcp.call is a write-class draft gated on canWrite.

Added

Measured variant outcomes feed back as learned opening-shape patterns that steer future generations, and a “Remember this rule” action on any quick edit pins your standing instructions where every AI function reads them.

Added

Notey could only talk when talked to. Now it can open the conversation: once a day it sends a Telegram digest of what it sees coming — notes scheduled in the next 24 hours, anything that failed to publish, drafts still waiting on you, evergreen pieces that have never gone out, and subscriber counts but only when they’ve actually been measured. When there’s genuinely nothing to report, it says so instead of inventing filler. It reports and suggests — it never queues or publishes anything on its own.

Added

One team, one Notey: its persona now comes from a single place and rides into every conversation. The built-in default writes like a content native — match your voice first, cut filler, treat cadence nudges as a service rather than a quota. In the AI tab’s Context section you can replace it wholesale with your own soul file, and your text wins exactly as written. Two rules survive any rewrite: Notey never invents facts, and nothing publishes without your explicit approval.

Added

When the model behind Notey got saturated, requests used to fail outright after three instant retries inside the same jam. Now Notey waits out short bursts with spaced retries first, and only if the configured model stays unreachable does it fall through to provably free alternates from the same provider, with a hard cap on how far down the chain it will go. When a swap happened, the reply tells you which model actually answered — never swapped silently mid-conversation — and a turn that already ran tools is never replayed in a way that could duplicate drafts or spend.

Added

“Who am I online?” and “how big is my audience?” now get live answers, not guesses. Notey pulls your current handle, display name, bio, avatar and follower count straight from each platform’s own API at ask time (Substack, Bluesky, Mastodon), and aggregates audience numbers — followers, following, posts — from the same verified self-lookups. Platforms with no honest pipe stay out entirely: X returns an explicit refusal rather than scraped numbers, and Substack audience totals point you at its own dashboard instead of inventing a figure.

Fixed

Talking to Notey over Telegram could feel broken: a thoughtful answer sometimes took longer than Telegram was willing to wait, so the message looked like it vanished — or came through twice when the delivery was retried. The heavy thinking now happens after the delivery is acknowledged instead of during it, so every message is received exactly once and the reply lands whenever the answer is ready. Your words are still saved first, before anything else runs.

Changed

Ask "who are you" in the Telegram chat or the in-app Notey panel and the answer used to be whatever raw model name was configured that day. Every conversation now opens from a fixed identity: Notey, NoteSurge's content agent, running on its current AI model (named for transparency, never claimed as self). Works identically on both surfaces because they share one agent loop.

Added

An MCP token used to be all-or-nothing: connect Claude Desktop or any other agent and it could do everything your connection allows, with no way to hand one client read-only access. Every token now carries tier checkboxes — Read (list notes, account status, best times), Draft (save drafts, library writes, repurpose), Publish (schedule real publishing) — and every call is checked against those tiers at a single gate before anything runs. A token without Publish checked can never schedule real publishing, no matter how it's prompted.

Added

NoteSurge's AI features have always run on your own key, but until now the menu had exactly two items: Anthropic or OpenAI. OpenRouter is now a third — bring one key from openrouter.ai and reach hundreds of models (Claude, GPT, Gemini, Llama, Mistral and more) under a single login and bill. Connect any two providers and choose which one is active; everything from rewrites and hook variants to Notey's chat follows your pick.

Added

Notey used to live only in Telegram: to ask about your queue or request a draft you left the app and opened a chat there. Now the AI tab has its own chat panel, wired to the very same assistant and the very same conversation history — ask something here and your Telegram bot sees it too; reply from your phone later and it shows up here. Nothing new to set up: if Notey has a model key it works immediately, with or without the Telegram connection.

Added

Ask Notey for something like "turn this essay into a week of content" or "do an archive dive — five fresh Notes from stuff I already wrote," and it now runs the whole play as a named skill instead of improvising call by call. Before anything happens, Notey shows you the plan — every step it's about to take — and only runs it once you're good with it. Three skills ship today: essay-to-week-of-content, winner-to-carousel-and-thread, and archive-dive-fresh-notes.

Added

Generation used to know your voice but not your track record. Now the AI that repurposes posts, writes hook variants and brainstorms ideas also reads the evidence from your own published notes — which themes and posting shapes actually earned engagement from your audience.

It arrives as a quiet evidence note inside the prompt, not a rule: "your hiring posts outperform" steers tone where it fits naturally, while what gets picked stays ranking's job. Editing passes like fix-grammar stay completely evidence-free.

Added

There's a new "Standing instructions" section in the AI tab. Type a rule once — "no em dashes", "never start with a question" — and every rewrite, adaptation, quick edit, and agent reply follows it. No more re-typing the same correction every time.

You can add or remove instructions any time, and everything listed there is something you typed on purpose — nothing is remembered silently.

Added

Any piece in your Library can now be pinned as a voice reference. Adapt-for-LinkedIn and every other generation leans harder on that piece's style when you do.

Pins show up in the AI tab next to your profile, and you can unpin from either place. Up to 3 at a time.

Changed

"Adapt for LinkedIn/Facebook/Mastodon" used to reshape your text with no idea who you are — none of your profile, voice, or rules reached it. That's fixed: it now follows your standing instructions and reference pieces like everything else.

Added

Anthropic was the only option before this. Now you can connect an OpenAI key instead, or alongside it -- rewrite, hook variants, Note ideas, repurposing, and the Telegram agent all work with either.

Connect both and a switch appears so you can pick which one is actually doing the work. Each provider also gets its own optional model field if you want something other than the default.

Added

The MCP & API section now lists every tool an agent gets once you connect one -- what each one does, and whether scheduling is locked until you turn on "Allow agents to schedule/publish."

A new get_account_status tool also lets an agent check what's actually connected before trying to draft or schedule anything, instead of finding out by trial and error.

Changed

Profile is gone, but nothing in it disappeared. It moved into a new AI tab as the "Context" section, right next to three things that used to be buried inside Settings: your Anthropic key, the conversational Telegram agent's on/off switches, and the tokens that let an outside AI agent (Claude Desktop, Claude Code, or anything else that speaks MCP) connect to your account.

One tab, four sub-sections: Context, Agent, Models & keys, MCP & API.

Added

Every image you upload in Compose or Requeue has a "Generate alt text" button under the thumbnail now. Click it and NoteSurge sends the image to your own Anthropic key and comes back with a one- or two-sentence description you can edit before using it.

Added

There’s a new Profile tab, and it’s not a Settings subsection pretending to matter. It holds four things: your bio, your goals, who you’re actually writing for, and any notes you want to leave about your own voice.

Fill it in once and it feeds the two AI transforms that create genuinely new material: hook variants and Note ideas. AI Quick Edits and the rewrite/newsletter/essay transforms don’t touch it, because those are editing text you already wrote in your own voice. There’s nothing for a profile to add there.

Added

Select some text in Compose, or select nothing to grab the whole draft, and you can rewrite it without leaving the page: fix grammar, improve it, condense it, make it punchier, or type your own instruction.

`⌘/Ctrl+J` opens the panel from anywhere in Compose, and every edit can be undone with one click if it’s not what you wanted.

Added

Three new AI transforms landed on library pieces.

Added

Message your own Telegram bot and it writes back like a co-writer, not a form. Riff on an idea, ask it to draft a Note in your own voice (it learns that from your last 15 published notes), or let it ask a few sharp questions when you’re stuck and don’t know what you’re trying to say yet.

Every reply has a save button under it that drops the text straight into your Library as an editable idea.

Changed

The "repurpose today" ranking stopped scoring purely on a post’s original X engagement, which by now might be years old and telling you nothing useful.

It adjusts each candidate by how similar content has actually performed once it became a Substack Note, and it says plainly which signal it’s using, so you’re never trusting a number more than you should.

Added

Any MCP-compatible AI agent, Claude included, can talk to NoteSurge directly now. Three tools: list your notes, draft one (always safe, it never auto-publishes), or schedule one, but only if you’ve turned on publish access for that specific token.

Added

Drop your own Anthropic API key into Settings and two buttons show up: "Polish with AI," which turns a raw imported tweet into an actual Substack Note, and "Expand with AI," which grows it out into a full newsletter draft.

Both are editable before you commit to anything.