Known breaks in Hermes Agent from github.com/NousResearch/hermes-agent, with quoted workarounds and fixed-in versions where they exist.
Type the version you're about to install (or just upgraded to). Lists tracked issues that mention it. Same data as /api/issues/hermes.json.
Pulled from public GitHub issues updated in the last 45 days and filtered for always-on problems: install and update breakage, crashes and hangs, gateway/daemon trouble, auth, and headless setups. Ranked by engagement, weighted toward always-on, self-hosting, install/update, crash, auth, MCP and daemon problems (cosmetic desktop-UI issues are down-weighted). Workarounds and "fixed in" versions are quoted only when the issue or its top comments state them. Severity is an automatic estimate from labels, keywords, and engagement, not an official rating. Last refresh: Oct 26, 2026 6:18 AM ET.
I want to use a remote Hermes Agent instance while keeping tool execution on my local machine. Machine A: local client / workspace machine Machine B: remote Hermes Agent with my existing skills, memory, sessions, and model/provider config…
What changed. On 14 October, #106742 (the unified gateway) moved every local conversation into one gateway process. The CLI, the terminal UI, the local Desktop app, the API server, editors, messaging bots, cron jobs, Kanban cards and Group…
The only workaround is config (`ws_orphan_reap_grace_s=0` plus a higher `HERMES_TUI_SESSION_TTL_S`). "I expected orchestration to continue with the laptop off." - [https://discord.com/channels/1053877538025386074/1563120848624164905](https://
Multiplayer agentic AI is the future. Hermes can and should lead. TL;DR: Memory operations bypass the hook system entirely, making tenant isolation impossible without forking core — we've been running a fix in production for months with…
A /model pick made in any gateway chat (Telegram, Bot Chat, TUI) on a session that follows the profile config fails with an internal error ("Model switch failed") whenever the session policy carries private config secrets…
I'd like to contribute a Nextcloud Talk messaging-platform adapter for the gateway — the same shape as the existing Mattermost / Signal / Telegram adapters: Hermes both receives messages from and sends replies to a Talk conversation, fully…
On startup, this warning is printed (twice in my case) before the agent begins processing: The messages render over already-drawn TUI content, garbling the screen. After the warnings, processing starts and works normally — solstice is a…
After updating past slack-sdk 3.44.0, the Slack working-status indicator ("is thinking…", live per-tool status via livestatus / typingstatustext) silently disappears on every turn. All status calls fail with invalidarguments; status…
Today, Hermes Agent supports exactly one terminal backend at a time — set globally via terminal.backend in config.yaml (mapped to TERMINALENV). You pick local, ssh, docker, modal, daytona, or singularity, and every terminal() call routes…
With discord.autothread: true, every automation wake that targets a session whose thread was auto-created is refused at admission: The drain requeues the event and retries forever, so parent agents never resume after their children finish…
Plugin tools can't raise their own execution deadline. PluginContext.registertool() (hermescli/plugins.py) does not accept or forward executiontimeout, but tools.registry.register() supports it and…
`timeouts.tools.sequential_call: 960` in `config.yaml`. This raises the ceiling for every tool, so a wedged tool now blocks for up to 16 minutes instead of 7. It also leaves the concurrent-batch path at 420 s.
On v0.21.6 the version line still carries the release date of v0.21.5: The version is right (it comes from the install stamp), but the date in parentheses is hermescli.releasedate, which is still "2026.9.24" in the v0.21.6 tag and on main…
Hermes v0.14.0 (latest, updated 2026-05-27 with 471 commits pulled) crashes on every hermes chat call when using openai-codex provider with gpt-5.5. The ChatGPT Codex backend (chatgpt.com/backend-api/codex) returns output: null in…
A single transient 429 / modelentitlement cooldown on one credential benches that credential (and often the whole provider row, or every auxiliary call) for hours to days, with no reset-time probe, no visible "in cooldown until T because…
When hermes update / the Desktop updater fails, the install is left half-applied (Desktop launches then closes, gateway venv missing pydanticcore, git checkout carrying partial-clone config) and the product prints the raw underlying error…
In a forum-enabled bot DM (Telegram "Threads" feature), when the user creates a new topic via the client's "New thread" composer, Telegram now delivers the opening message without messagethreadid (thread-less), while the forumtopiccreated…
Hi! I'm developing a skill that uses the following stack: Kimi K3; from OpenCode Go Bitwarden; for secrets (using the "official" resource from Hermes docs) Browserbase headless; integrated in Hermes (free) without API key - as is. Hermes…
Telegram init hangs pre-socket inside the gateway process after a routine network blip, while the identical stack (same venv, same token, same transports, same SSL context) connects in 0.3s standalone. Appears to be a new variant of the…
On Linux, launching Hermes Desktop again while it is already running (the launcher, gtk-launch, or a hermes:// deeplink that starts the binary) can leave windows-sandbox-fallback.json stuck at {"state":"fallback","reason":"boot-loop"}…
What happens: hermes -z "…" (and any entrypoint that ensures the gateway) fails with Error: Gateway inaccessible: servicemanagerunavailable when run without a GUI login session — e.g. from a system LaunchDaemon with UserName= , or via sudo…
hermes kanban swarm's verifier role hardcodes the skill requesting-code-review (hermescli/kanbanswarm.py:238). This skill ships bundled (skills/software-development/requesting-code-review/), so it normally exists — but the Curator's…
On a self-managed (shell-installer / PM) Hermes install, every scheduled cron job fails before it can run. The restart-safe external worker is spawned as sys.executable -m cron.scheduler with a sanitized environment whose PYTHONPATH…
A Windows install that moved to the PM runtime (managed Python 3.14, dependencies under installs/ /environments/ /venv) still has the old pre-PM \venv (Python 3.11) on disk. After hermes update --gateway respawned the messaging gateway…
pm/workspace.py::membersstamp() hashes every file in each enabled plugin member directory to decide whether the recorded dependency environment is current (pm/install.py::venviscurrent). The only exclusions are the hard-coded MEMBEREXCLUDE…
A git-installed Hermes update on Ubuntu 24.04 failed while preparing the new isolated dependency environment during the historical PM takeover. The selected features included matrix; mautrix[encryption] pulled python-olm==3.2.16, which…
Opening a conversation that belongs to a messaging platform (Telegram DM/group, WeChat, Feishu, …) in Hermes Desktop leaves the composer's reasoning-effort pill in its loading state indefinitely. It never paints the chat's actual level…
After upgrading to an environment that has httpx2==2.12.0 (the pin Hermes carries for the mcp extra since bc8545bd1a, "bump httpx2/httpcore2 2.7.0 -> 2.12.0 for the audit advisories"), any HTTP MCP server whose tools/list response arrives…
The gateway-owned group chat workspace (canonical-group, Desktop v0.22.2) lost several capabilities the legacy Desktop room UI had. Five gaps, all in the same surface: Composer: no clipboard image paste, no drag-and-drop — a bare in…
After updating a git/source installation to Hermes Agent v0.21.5+2959.ga342e65 (commit a342e6563d), the automatic migration of the Hindsight memory provider from core to the plugin catalog gets stuck while preparing the plugin's Python…
On macOS, every update started from the Desktop app's Update button fails with exit code 2 because hermes update refuses the lock held by its own hand-off: The PID named in the error is the custodian that scripts/desktop-update/posix.sh…
quit the Desktop app and run `~/.hermes/hermes-agent/.hermes/bin/hermes update` in a terminal.
On the official Docker image (linux-arm64, v0.21.6) with HERMESHOME on a case-insensitive volume (macOS/APFS default — a plain Docker Desktop bind mount under /Users/...), no dependency environment can ever be committed for the install…
hermescli/kanbandbdispatch.py::pidrecycled has a string branch for the composed worker fingerprint (" | ") that compares the recorded value to a freshly computed one with EXACT equality. On hosts without /proc (macOS) the start-time half…
resolvehermesargv() in hermescli/kanbandbdispatch.py decides the kanban worker's spawn argv using an importability check evaluated in the gateway (parent) process, but that result does not hold for the freshly-spawned child. On installs…
Pin `HERMES_BIN` to the shim for the gateway via a systemd drop-in (survives self-update, unlike a source patch):
Hermes Agent intermittently fails with APIError and servererror when using the openai-codex provider with gpt-6.1-sol. The issue occurs in both Hermes Desktop and Discord Gateway on Windows 11. Some requests complete successfully…
In Telegram DM topic mode, when a topic is actively running, typing a new message in the All lane to start a new topic: creates a new (empty, dead) topic, and the message text lands in the actively running topic/session instead — the new…
Hermes Agent: v0.21.3 Server: macOS (Mac mini) Surfaces: Desktop app + mobile dashboard When a session is open on the Desktop app (surface=desktop), a second surface (mobile dashboard, surface=tui) cannot submit messages to that same…
tools.lazydeps.installspecs is now a shim that calls hermescli.oldupdater.stopforrelaunch(). That function starts the full historical update takeover (updatetakeover.py → updatefinish.py), which syncs PM and rebuilds Node deps, the TUI…
Setting `memory.provider: ''` stops the loop. After Desktop relaunched, agent init finished normally and no takeover started. Hindsight memory is unavailable while this setting is in place.
The nousresearch/hermes-agent:latest Docker image on Docker Hub does not include the dashboard (or web) subcommand, making it impossible to run the web UI in containerized environments like Coolify. The dashboard command should be…
Hi i tried to update the hermes client on MacOS but it failed. Error message is only Update failed so no info there :). Hermes Agent v0.21.5+8076.g85db7c3 (2026.9.24) · upstream 85db7c3a Install directory…
After updating to the latest main (24fd22b94d, v0.21.3 / 2026.9.14), the interactive CLI crashes immediately on startup with: This is a regression from 182ec5c28d (#111408, "widen the remaining config.yaml stat caches to filesignature")…
Since #127201 (d428df2a65, "fix(desktop): restore YouTube embeds in the packaged app"), the packaged Desktop renderer is served from http://127.0.0.1:47891. Its WebSocket upgrades therefore carry Origin: http://127.0.0.1:47891. When…
No issues match those filters.