Twitxy Saved

Hermes Release Watch

Verified account This account is verified. Learn more

@HermesWatcher

Tracking every Hermes Agent release, changelog & major update — plus what actually matters.

Unofficial. Not affiliated with Nous Research.
hermeswatcher.com/links Joined July 2026
885 Following 29K Followers 9.6K Posts

Latest posts 5

Pinned post

Hermes Release Watch Verified account @HermesWatcher
Every Hermes user needs to see this.

If you run Hermes, real long-term memory is a must. Your agent should know you, not make you repeat yourself.

Hermes already keeps notes on you. Honcho takes it further.

So I put it to the test (new accounts get $10 in credits to try it) 👇

Honcho is an open-source memory layer for AI agents. It runs alongside Hermes's own memory and builds a long-term picture of you in the background.

The test:
▸ Told Hermes on Telegram I'm leading the Q4 pricing launch. Later: "Heads up, standup moved to 10."
▸ Opened a brand-new Hermes chat in the terminal and asked it to plan Monday morning. Standup at 10:00.
▸ Switched to Claude Code, a completely different agent, and asked it to draft my standup update. Written for the 10:00 standup, about the pricing launch.
▸ Opened the Honcho dashboard to see what it concluded, shown next to the Telegram message it came from.

Said once, on Telegram. Every agent picked it up.

What else it does:
▸ Learns your preferences, goals and how you work
▸ Draws conclusions from patterns across your conversations
▸ Keeps up when things change (a correction becomes a newer memory)
▸ Works with Hermes, Claude Code, Codex, Cursor, OpenClaw and more
▸ Open source: self-host it or use Honcho Cloud

This is the memory setup I'd tell every Hermes user to add.

New to Honcho? Your $10 in credits is waiting in the first reply 👇
1:40
6 replies 7 reposts 133 likes 15K views View on X
Hermes Release Watch Verified account @HermesWatcher
Built your Hermes agent locally? You can now transfer the whole setup to Hermes Cloud, right from Desktop. 👀

54 PRs merged on November 16.

4 changes worth knowing:

▸ Backup & Transfer is now built into Desktop Settings. Back up your entire setup or replace an existing Hermes Cloud agent with your local one, including memory, skills, conversations, profiles, and scheduled jobs. There's even an Undo option

▸ Rapid-MLX joins the plugin catalog, bringing managed local AI models to Apple Silicon Macs with downloads, model selection, and runtime controls

▸ OpenDocKit lets Hermes read PDFs, Word documents, spreadsheets, presentations, and more as Markdown, including individual pages, slides, or sheets

▸ Telegram can now show reasoning above the final answer in one message instead of sending two separate replies

Luke has the full 54 below 👇
Luke The Dev Verified account @iamlukethedev 3h
Hermes merged 54 PRs on November 16

The ones worth knowing:

- Desktop: back up your agent, or replace a Hermes Cloud agent with it, from Settings
- Desktop cold-start resume keeps your caret and draft, and keys typed mid-remount survive
- Desktop reconnect pages now draw a stored reply once, in full
- Desktop SSH connect and forward timeouts now come from config.yaml
- Telegram shows reasoning above the answer, in one message
- Cloud answers "not signed in to Nous Portal" with a 409 instead of a 401
- Renaming a compressed chat from a stale tab now sticks, and the CLI status bar keeps the chat name after compression
- A Desktop reply streamed before its last tool settles now lands correctly (0.23.2 blocker)
- Approval floors are now heredoc-aware and flag fetch-then-run, pipe sinks, and brace command words
- New plugin catalog entries: managed Rapid-MLX runtime and opendockit, plus loredocs 0.1.34
- Big dead-code cleanup across MCP, delegation, browser, desktop, skills, and approval
- Windows CI lane no longer times out replaying workflows

Here is the full list:

DESKTOP IMPROVEMENTS (7 PRs)

- fix(desktop): the cold-start resume keeps the caret after the carried draft and keeps keys typed mid-remount
- fix(desktop): keep the boot-time draft when the cold-start resume remounts the composer
- fix(desktop): a reconnect page that stores the reply draws it once, in full
- fix(gateway): a Desktop reply streamed before its last tool settles again (0.23.2 blocker)
- fix(gateway): an adopted Desktop row with bare custom resumes on the profile route
- Desktop: back up your agent, or replace a Hermes Cloud agent with it, from Settings
- fix(desktop): take the SSH connect and forward timeouts from config.yaml

VOICE & REAL-TIME (2 PRs)

- refactor(speech): one TTS extras table, no STT lease counter, no import-time output dir
- refactor(speech): drop the unused stt_config parameter of _is_local_stt_provider

DEVELOPER TOOLS & PLUGINS (16 PRs)

- refactor(skills): stop swapping SKILLS_DIR in the dashboard profile scope
- refactor(environments-terminal): per-call process checkpoint path, drop test-only aliases
- refactor(delegation): siblings read config from its owner; drop status words no writer emits
- refactor(connectors-mcp): drop MCP loop context wrappers, MCPServerTask slots and MUTATING_TOOL_NAMES
- refactor(code-files): resolve checkpoint and write-guard paths per call
- refactor(browser-desktop): move test hooks out of computer_use and bot_desktop
- refactor(core): drop the stale toolset-check map and TOOL_TO_TOOLSET_MAP
- refactor(skills): delete unused hub aliases and the blueprint export functions
- refactor(environments-terminal): delete unused sandbox helpers and the failed_start status
- refactor(delegation): delete an unused ledger ack, reserved events and dead parameters
- refactor(connectors-mcp): delete unreached connector and MCP OAuth code
- refactor(code-files): delete an always-true sandbox gate and unused checkpoint helpers
- refactor(browser-desktop): delete unreached browser and desktop tool code
- feat(plugins): catalog managed Rapid-MLX runtime
- plugin-catalog: bump loredocs to 0.1.34
- feat(plugin-catalog): add opendockit plugin

SECURITY & SECRETS (3 PRs)

- refactor(approval): one pending-approval lookup; an approval entry starts Deny-only
- refactor(approval): delete an unread pending store and redundant guards
- fix(approval): heredoc-aware command floors; flag same-string fetch-then-run, group/interpreter pipe sinks, brace command words

INTEGRATIONS & CHAT (6 PRs)

- Telegram: reasoning shown above the answer, in one message
- fix(gateway): a word holding a key prefix mid-word is not held off the streamed preview
- fix(cloud): answer 'not signed in to Nous Portal' with 409, not 401
- refactor(gateway): outbound send files move from tools/ to gateway/outbound/
- refactor(cron): cron helper files move from tools/ to cron/
- refactor(core): delete unreached send_message actions and unused tool helpers

STABILITY & FIXES (20 PRs)

- fix(ci): Windows lane no longer times out replaying workflows or races a killed backup's lock
- fix(sessions): renaming a compressed chat from a stale tab sticks instead of reverting
- fix(cli): the chat's name stays on the status bar after it compresses
- fix(update-check): the passive check's git ls-remote dies with the CLI
- test(e2e-upgrade): handoff tests the restart-pending report, not the removed cap
- test(install-e2e): tolerate pre-679e9cd2 in-checkout Windows launchers on Desktop-update legs
- test(e2e): handoff scenario lets the held kanban worker finish before waiting for the relaunch
- fix(install-e2e): a profile's cache/ is advisory, not user state
- fix(move): flag a path joined through a folder variable to a moved file
- fix(move): move scripts handle files that leave a folder that stays
- test(browser-desktop): pin desktop_preview close on the live path
- refactor(utils): split utils. py into a package of three modules
- refactor(tools): snapshot stores call read_json_or_empty
- refactor(utils): make the file mode and owner helpers public
- refactor(agent): move the max_completion_tokens model rule out of utils
- refactor: six JSON readers call read_json_or_empty
- refactor(utils): retry a contended replace without importing agent
- refactor(utils): drop the sys.modules worker-policy lookup
- docs: imports go at module top; drop the late-import advice
- fix(code-health): the file-size fix text asks for a folder, guarded by a discovery test

54 new features and fixes

x.com/iamlukethedev/…
View quoted post on X
Hermes Release Watch Verified account @HermesWatcher
MY AI DIDN'T MAKE THIS ONE UP.

I asked Hermes a question and it went and read the official sources.

It searched the web live, read the actual pages, and put a source on every single claim.

Click any source to open the page and check it yourself.

Hermes has had web search built in for a while, no search API key needed. Most people just never use it this way.

Try it for:
▸ Checking which plants are safe for your pets (that's the video)
▸ Comparing products with the spec pages attached
▸ Researching a trip, a recipe or a law with sources
▸ Building a work reading list with links

Your AI can show its homework now 🔥
0:44
5 replies 0 reposts 30 likes 3.2K views View on X
Hermes Release Watch Verified account @HermesWatcher
UPDATE HERMES TODAY. IT STOPS DROPPING YOUR WORK MID-JOB.

v0.23.2 is out, and the biggest fixes are the ones you've felt:

▸ Updating or restarting no longer leaves your bot ignoring you for up to 30 minutes, then killing what was running. It waits until the work is done
▸ Long builds and installs no longer die at exactly 7 minutes
▸ Helper agents cut off by a crash or restart can pick up where they stopped
▸ A huge answer in Telegram or Discord arrives as one file, not a wall of messages
▸ When a site asks for a login code, Hermes asks you for it in chat
▸ The security lockdown we told you about is now in a stable release

Run hermes update, or update in the Microsoft Store.