hermes: fall back to sference when hyper is unavailable #12

Open

hermes wants to merge 4 commits from hermes-fallback into main

pull from: hermes-fallback

merge into: repomaa:main

repomaa:main

repomaa:hermes-dashboard-auth

repomaa:fix/ci-deploy-polkit

repomaa:pr/hastebin-client-sops

repomaa:ci/fqdn-deploy-trigger

repomaa:hermes-dashboard

repomaa:feat/gatus-replaces-kuma

repomaa:fix/apu-wake-ifd

repomaa:chore/remove-stray-files

repomaa:chore/turny-use-nix-aspect

repomaa:chore/colmena-drop-dead-ssh-specialarg

repomaa:fix/mkenableoption-descriptions

repomaa:chore/dedupe-flake-file-inputs

repomaa:fix/dedupe-unstable-overlay

repomaa:forgejo-sse-buffering

repomaa:hermes-free-response-room

repomaa:hermes-rbw

repomaa:hermes-webhooks

repomaa:feat/dendritic-refactor

repomaa:tech/native-immich-module

repomaa:refactor

Conversation 0 Commits 4 Files changed 2 +16

hermes commented 2026-09-26 14:49:33 +03:00

Collaborator

Copy link

Sets fallback_providers to sference's zai-org/GLM-5.3-Flash so sessions keep running when the hyper subscription is out of credits or failing.

How the auto-return works (no cron/checker needed): Hermes fallback is turn-scoped — every new message retries the primary (hyper) first. If hyper is still out of credits, the turn runs on sference; once the subscription renews, the next message naturally succeeds on hyper. Result: 'automatically on hyper again' with zero intervention, and no staleness window from a scheduled checker.

Eval-verified: containers.hermes.config.services.hermes-agent.settings.fallback_providers = [{model: "zai-org/GLM-5.3-Flash", provider: "sference"}].

Sets `fallback_providers` to sference's `zai-org/GLM-5.3-Flash` so sessions keep running when the hyper subscription is out of credits or failing. How the auto-return works (no cron/checker needed): Hermes fallback is **turn-scoped** — every new message retries the primary (hyper) first. If hyper is still out of credits, the turn runs on sference; once the subscription renews, the next message naturally succeeds on hyper. Result: 'automatically on hyper again' with zero intervention, and no staleness window from a scheduled checker. Eval-verified: `containers.hermes.config.services.hermes-agent.settings.fallback_providers` = `[{model: "zai-org/GLM-5.3-Flash", provider: "sference"}]`.

hermes added 3 commits 2026-09-26 14:49:33 +03:00

hermes: add rbw + container tools to extraPackages ... 61c1f0ba6a

rbw now works end-to-end against the vault (pinentry stub fixed, account
re-registered with standard key wrapping). Include it with git, forgejo-cli
and jq as the agent's day-to-day toolset.

ci: trigger freun-dev deploy on merge ... 576f9e2f0f

Adds a Forgejo Actions workflow that, on push to main, ssh'es to freun-dev
as the restricted ci-deploy user. The account's authorized_keys entry is
pinned to a single command (systemctl start nixos-upgrade.service) with
port/X11/agent forwarding and PTY allocation disabled — CI can only kick
the existing pull-based upgrade, not run arbitrary commands. The upgrade
then fetches main via the existing read-only deploy key and switches.

Deployment flow on merge:
1. check.yml validates the flake (already in place)
2. deploy.yml triggers nixos-upgrade on freun-dev
3. the upgrade unit fetches the new revision and switches, with the
   existing kuma heartbeats reporting success/failure

Setup remaining (operator):
- sops-encrypt the private key: 'sops secrets/ci-deploy-key' with the
  plaintext from the agent (staged at the agent's secrets dir), then add
  its base64 as the CI_DEPLOY_KEY repository secret
- merge + apply freun-dev once to install the ci-deploy user

hermes: fall back to sference when hyper is unavailable ... 8ad043438a

fallback_providers routes sessions to sference's GLM-5.3-Flash when the
hyper subscription is out of credits or otherwise failing. Fallback is
turn-scoped: every new message retries hyper first, so a renewed
subscription is picked up automatically without manual intervention.

hermes added 1 commit 2026-09-27 22:30:58 +03:00

merge main into hermes-fallback ... b77afb3483

# Conflicts:
#	.forgejo/workflows/deploy.yml
#	modules/ci-deploy.nix

hermes force-pushed hermes-fallback from b77afb3483 to b3a5efeabc 2026-09-27 22:31:19 +03:00 Compare

repomaa force-pushed hermes-fallback from b3a5efeabc to 47a1e79b11 2026-09-27 22:48:54 +03:00 Compare

repomaa force-pushed hermes-fallback from 47a1e79b11 to ab23193bfd 2026-09-27 22:56:27 +03:00 Compare

All checks were successful

Hide checks

Build Images / build (pull_request) Successful in 41s

Details

Check / check (pull_request) Successful in 3m50s

Required

Details

This pull request has changes conflicting with the target branch.

• modules/hermes.nix

View command line instructions

Manual merge helper

Use this merge commit message when completing the merge manually.

Merge commit title hermes: fall back to sference when hyper is unavailable (#12)

Merge commit body Reviewed-on: https://git.freun.dev/repomaa/nixos/pulls/12

hermes: fall back to sference when hyper is unavailable (#12) Reviewed-on: https://git.freun.dev/repomaa/nixos/pulls/12 Copy merge commit message

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u origin hermes-fallback:hermes-fallback

git switch hermes-fallback

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

repomaa/nixos!12

No description provided.