README.md

• Voice — voice, audio and video notes transcribed with Deepgram nova-3; auto-detects ru/uz/en.
• Vision — photos described by your provider's own vision model; no extra key, no extra bill.
• Rich replies — tables, checklists, collapsible blocks and formulas render natively in Telegram via Bot API 10.1 rich messages; plain formatting keeps its proven path, with a graceful fallback.
• Quiet update checks — once a day Iva checks for a newer stable release without spending model tokens. If one exists, Telegram offers Update or Later once; otherwise it says nothing.
• Layered memory — remembers across months, long after the chat window has scrolled away.
• Personal CRM — who your people are, what you agreed, when to follow up.
• Search by meaning — BM25 plus link-graph rerank, any language; optional vector mode with one key.
• Decision cards — what you chose, when and why; old versions stay in a dated History.
• Tasks & reminders — priorities, due dates and a morning digest.
• Web search — four pluggable providers: Tavily, Exa, Parallel or Brave.
• Google Workspace — Gmail, Calendar, Drive, Sheets, Docs and Tasks from chat via the gws CLI; installed for you, with a guided key setup right in the conversation.
• Skills & MCP — drop one file to add a procedure or connect an MCP server; keys stay in .env.
• Personal Telegram — userbot (beta) — read and send from your own account, not just the bot; connect by chat (QR, no terminal). Rough and buggy — opt-in, at your own risk. A server-side anti-ban guardrail (FloodWait compliance + randomized pacing + circuit-breaker) is enforced, not just advised. Details.
• Safe to forward — forwarded text, captions and voice transcripts pass an injection screen before the model reads them. A flagged message or transcript reaches the model tagged as data rather than as an instruction; for media captions the screen runs but the tag does not travel with it yet.
• Token accounting — every model step is logged; /usage reports it for free.

git clone https://github.com/smixs/iva-agent.git ~/iva
cd ~/iva && bash install.sh

The installer reuses the existing checkout instead of re-cloning, keeps .env and the vault untouched, and installs the same dependencies. A fork or a branch works through variables read at startup: REPO_URL=…, BRANCH=…, INSTALL_DIR=… (defaults: this repo, main, ~/iva). Details: docs/install.md.

17.09.2026

v0.4.4

• ⏰ A reminder is an instruction Iva gives her future self: at the due minute the text of the reminder is the prompt of one fresh turn with tools, not a line to read back — "in 3 minutes find the news and send it" arrives as the news, not as its own wording. A turn that could not run, failed or came back empty still delivers: the code sends your text verbatim and names the cause in the row.
• ⏰ A reminder comes back to the chat it was asked in: the row remembers the chat and the topic of the turn, so a reminder set in a group topic lands in that topic instead of a private chat; old rows and requests that did not come from Telegram still go to the owner's chat. iva doctor stopped claiming "the dispatcher has not ticked yet" — it was reading the pulse from the wrong folder.
• 🔁 /update --force works from the chat too: the word after /update is read now, so the running build can be rebuilt from Telegram exactly as iva update --force does on the server — the way out of a broken edit in Iva's own code. The flag travels in the request file, so a restarted update rebuilds as well.
• 🩺 The doctor names the cure for stuck workflows: "running count 7 exceeds 5" now says in the same line that such runs are stale rather than live, and that iva reset quarantines them and restarts the services; memory is left alone.

14.09.2026

v0.4.3

• 🔁 One updater, the way pi does it: the old in-place update path (stash and rebase inside the working folder with a byte-level check of stray files, ~15k lines) is gone together with the guesswork "developer or installation" by branches and shims that kept people with a second branch on the fragile path forever. Every Iva folder now updates through versions: build beside, probe, switch, roll back. A developer checkout is marked with an empty .iva-dev file. Edits to Iva's own code are no longer promised or kept — your own skills, tools and plugins live in data/custom. repair.sh and a re-run of install.sh hand an existing installation to the same updater. An update cut off mid-way (a server reboot) is restarted once by the bridge itself with a line in the chat. Older flat installs need two /updates: the first fetches the new code, the second moves onto versions.
• 🔘 Menu buttons two per row again: the classic menu lays buttons out in pairs as before 0.4.2; the new rich menu shows the same pairs as compact pills with a "button — what it does" caption under the row, no more one long full-width button.
• 🔌 A tool schema the provider rejects no longer kills the turn: OpenAI (codex) rejects the whole request when any tool carries a regex with lookaround; Iva now retries once without those patterns, and if it still fails, the error names the field and where the tool lives.
• 🧷 Codex tools without strict mode: tools go to codex with strict: false, so optional fields stay optional and reminders are set on the first call instead of looping.
• 🧰 diagnose.sh collects more: the plugin list, the reminder dispatcher pulse and the schedule lines of the last day.

13.09.2026

v0.4.2

• 🎤 The voice key is no longer required at install: the Deepgram step of the wizard is skipped with Enter (console.deepgram.com does not open from some countries, and people got stuck on install for an optional feature); without the key voice notes are saved and Iva suggests /menu → 🎤 Voice, and iva doctor shows a warning instead of a failure.
• 🔘 A new menu, on request: buttons inside the message: by default the menu, the /model and /think wizards, the update offer and the "Working" status look as before (a message with buttons under it); /menu → Maintenance → ✨ New menu switches them to Telegram rich messages — every button a full-width row with what it does right under it, headings, status and timers as tables, and ◀︎ Classic menu at the bottom brings the old look back. Iva also offers buttons in her own replies when there are two to four options, and a tap comes back to her as your message; the rich-replies skill covers the whole palette. The new menu needs a Telegram client from August 2026; in groups the reply buttons don't work. ADR-0015
• 💬 Rich replies and voice are set from the menu: two new /menu screens — a switch for rich replies (Auto: tables, task lists, folds and formulas go as rich messages; Plain text: ordinary messages) and 🎤 Voice with the Deepgram key and the recognition language (Auto/Русский/English/Oʻzbek); the key is taken from your next message in a private chat and deleted from the chat, and both screens offer a restart after saving.
• 📦 Dependencies cleaned up: the unused @vercel/connect is gone, fast-uri, hono and qs under @modelcontextprotocol/sdk are bumped past their vulnerabilities (npm audit: 0 high, 0 moderate), and the ai pin is lifted to ^7.0.82, the peer eve 0.51.1 requires.

v0.4.1

• ⏰ A reminder is an instruction Iva gives her future self: a data/ row moves "pending → fired" in one atomic transition, and at the due minute the text of the reminder is the prompt of one fresh turn — Iva does what it says with her tools and the code sends the final text of that turn back to the chat and topic you asked in, so "in 3 minutes find the news and send it" arrives as the news, not as its own wording. A turn that could not run, failed or came back empty still delivers: the code sends your text verbatim and names the cause in the row. The firing fact (fired_at, delivered, error) sits in the row, iva doctor shows it, and rows from the old schema survive the update.
• 🔔 One remind tool with add, list, remove: one-off reminders ("in 30 minutes", "at 14:30") and repeating ones (a cron expression in your time zone) are set, listed and removed by a single tool with an action field instead of three. The moment is computed in code and handed to the agent as a ready time, schedules firing more often than every 10 minutes are refused, and the destination is always your chat. ADR-0013
• 🛡 bash cannot set timers of its own or talk to Telegram anymore: systemd-run, writing crontab, at/batch, units under ~/.config/systemd/user, ~/.iva-scripts, sleep chains and direct calls to api.telegram.org are refused before they run, and the refusal names the replacement; reading (crontab -l, systemctl status, journalctl) still passes.
• 📋 Every schedule run leaves a fact, and the agent wakes up with it: each run writes a row to data/jobs.json (reason, exit code, secret-free tail, kept seven days). On success Iva stays silent; on failure she fixes the cause and tells the owner, open failures are visible to every turn and in iva doctor and close on the next success or with iva jobs ack <name>, and if the agent cannot wake at all, one message a day reaches you. A wake turn that ended by waiting for the next message is a normal end now, not a failure. The schedules section of iva doctor works without systemd.
• 🔎 A complaint turns into an evidence bundle with no secrets in it: iva diagnose puts versions, OS and node, the iva doctor output, the last 200 service log lines, reminder and turn-failure facts and the schedule table into data/diagnose/<date>.md, cutting the values of every .env key except settings, plus the bot token, owner chat id and e-mail; the report-problem skill reads the bundle, explains the failure in two lines and offers a ready issue link or a message for the support group.
• 🧹 An interrupted update cleans up after itself and names the reason: the retiring checkout wears a marker with its own identity, deletes .git last and finishes on a repeat; stale shim-refresh claims in ~/.local/bin are swept by age even under a live pid; a failed optional step now logs its exit code and the last output line (for example exit 127: uv: command not found) instead of a silent "… did not run".
• 🧩 Your rules live beside the bundled persona and load every turn: markdown files in data/custom/agent/instructions/ reach the prompt live without a rebuild, so a behavior rule written into rules.md (with write_file, after your confirmation) works at once; the old instructions.md replacement is obsolete, and iva doctor shows the rule count and warns when it overruns the limit.
• 🧭 The vault directory is computed by one formula everywhere: file tools, media, the diary, nightly memory, the CLI, the menu, the build and the installer call a single resolver. An empty value or stray spaces is now a clear error naming the variable instead of a silent directory swap, a relative path resolves against the caller's base, and a bad setting reaches the user as one line, not a stack trace.
• 🧩 Iva runs on eve 0.51.1 (0.4.0 shipped 0.47.3): the local patch is ported and rebuilt for the new runtime.