A follow-up sent during an active run now offers itself to the running
agentic loop via POST /api/ai/sessions/{id}/steer. Accepted steers join
the loop at its next turn boundary (the abort-check site) as plain user
messages, are announced with a steer_applied stream event so the drawer
settles the pending row, and persist through the end-of-run save. A
steer carries prompt text only: no route, control-level, or autonomy
changes, no turn-budget extension, system sessions rejected, and the
per-session inbox is bounded (steer_backlog overflow). Delivery is not
guaranteed by acceptance: a run that ends first discards the inbox and
the row drains as an ordinary queued turn, so pre-steering queue
semantics remain the fallback. Steering rows lose edit/remove once
accepted.
Chat turns, compaction, and title calls now stamp session_id on their
cost.UsageEvent, and the done event carries session_cost_usd summed from
the operator ledger (cost.Store.SessionCostUSD). The drawer's last-turn
summary appends '$0.12 session' with a sub-cent floor. The figure is
omitted whenever any of the session's models has unknown pricing, and
free local models price known-at-zero, so no figure is ever partial.
Pasting a log or config into the composer flooded the textarea. Pastes
over 8 lines or 800 chars now collapse into a removable chip; the full
text rejoins the prompt after the typed question at send time. Clicking
the chip body expands it back into the textarea for inline editing, and
a chip-only send is allowed.
The undo flow already restores the removed prompt into the composer for
editing, but the only entry points were the header undo button and /undo.
Surface it where users look for it: a hover pencil on the latest
non-queued user prompt that undoes the turn and prefills the composer,
riding the same last-turn undo boundary as regenerate.
The MCP server and its Settings setup panel existed but the README never
mentioned them, so users who prefer their own agent (Claude Code,
OpenCode) had no pointer to the supported path.
Retrying a failed turn re-sent the prompt without removing the persisted
turn, so session history double-recorded the prompt. Session undo now
accepts an expected-prompt guard (a stale retry can never remove a
different turn); retry drops the replaced turn server-side before
re-sending. The latest settled assistant answer gains a hover-revealed
Regenerate button that reuses the same path.
The infrastructure resource drawer saved guest metadata (web interface
URLs) under the unified resource hash or the discovery resource id, a
keyspace no workloads table reads: the tables and the workloads drawer
resolve metadata by the canonical id (instance:node:vmid for PVE guests,
resource id otherwise), which is also the key v5 upgrades carry over in
guest_metadata.json. A URL saved from that drawer was stranded where
only the same drawer could read it back.
Route the drawer's guest metadata id through
getCanonicalWorkloadIdForResource, and canonicalize the resource type
inside that helper so lxc/oci-container/qemu spellings build the same
node-scoped id the workloads surfaces use.
Refs #1556
The infrastructure resource drawer saved guest metadata (web interface
URLs) under the unified resource hash or the discovery resource id, a
keyspace no workloads table reads: the tables and the workloads drawer
resolve metadata by the canonical id (instance:node:vmid for PVE guests,
resource id otherwise), which is also the key v5 upgrades carry over in
guest_metadata.json. A URL saved from that drawer was stranded where
only the same drawer could read it back.
Route the drawer's guest metadata id through
getCanonicalWorkloadIdForResource, and canonicalize the resource type
inside that helper so lxc/oci-container/qemu spellings build the same
node-scoped id the workloads surfaces use.
Refs #1556
One-click updates recreated containers from the raw inspected config, so
Docker rejected the create for any container using network_mode:
container:<id> (compose service:<x>) or host with "conflicting options:
hostname and the network mode", leaving the workload stopped under its
_pulse_backup_ name until the rollback rename restored it. Docker fills
Config.Hostname with the namespace owner's ID on such containers, so the
verbatim copy always tripped the daemon validation (verified against a
real Docker 29.5.2 daemon: verbatim create 409s, sanitized create
succeeds and the replacement starts in the shared namespace).
Strip the namespace-owned settings before ContainerCreate: hostname and
domainname for container:/host modes, plus exposed/published ports,
links, DNS options, and extra hosts for container: mode, all of which
the daemon derives from the owning namespace. Bridge and user-defined
networks are untouched.
Refs #1564
rcourtman