[feat] Durable cross-device agent session list (archive, ended, titles) - #5479
Merged
bekossy merged 14 commits intoJul 28, 2026
Merged
Conversation
A kill/archive soft-deletes the session_streams row, but the full unique constraint on (project_id, session_id) keeps the tombstone in the slot. On the next turn _start_turn read None, create hit the slot, and update matched nothing, leaving the durable row dead while Redis went alive — so a resumed session vanished from the list and fetch/heartbeat returned None. Detect the no-match update and revive that same row (unarchive + re-flag), preserving its id and header across the kill boundary.
Add querySessions (POST /sessions/query) + extend the stream schema with the
header (name/description) and lifecycle timestamps, so the durable session list
is reachable from the FE. Add projectSessionsQueryAtomFamily(appId), mirroring the
liveness query: one low-priority, focus/interval-revalidated query per agent,
scoped by references [{id: appId}] against the turns' workflow refs. Not yet wired
into the sidebar — the reconciler that folds this over the localStorage list follows.
Kill soft-deletes the stream row, so the default query (deleted_at IS NULL) drops killed-but-resumable sessions from the list — under STOP_KILLS_SESSION that's most of them. Add include_ended to SessionQuery/SessionStreamQuery so the durable list keeps ended sessions (deleted_at populated for the caller to mark them). Absence from the list then means genuinely hard-deleted, which the FE uses to prune its cache.
Fold the agent's /sessions/query result into sessionsByAppAtom: adopt sessions the server knows and we don't (cross-device, and recovers a wiped localStorage), enrich title/createdAt (a local user title wins), and drop a session the server dropped — server-confirmed before, absent now = hard-deleted elsewhere — never a purely-local optimistic one (tracked via a new serverKnown flag; also exempts it from husk pruning). Open tabs/active stay per-device. Gated on query success so a failed fetch never drops.
renameSession only wrote the local cache, so a title never left the browser. Add setSessionHeader (PUT /sessions/streams/header) and fire it from renameSession — optimistic local update, best-effort server write. Combined with the reconciler adopting the server name, a rename now round-trips across devices and survives a localStorage wipe (creates the stream row if the rename precedes the first run).
Delete was localStorage-only, so a delete never left the browser. Add deleteSession (DELETE /sessions/, the root hard-delete) and fire it from deleteSessionAtomFamily for server-known sessions only (a local husk has no row; the reconciler's own drop path never routes here, so no loop) — a delete now propagates to every device. And surface killed/ended sessions: include_ended already lists soft-deleted rows, so carry an 'ended' flag (deleted_at) through the reconciler and show a muted 'Ended' chip in the rail and history popover. (Kill and archive both set deleted_at on this backend, so an archived session also reads as ended until the backend distinguishes them.)
Archive and kill both set deleted_at, so the durable list couldn't tell a killed-but- resumable session from a deliberately-hidden one. Add an archived_at column (migration 018) and repoint archive/unarchive to it: archive sets archived_at, unarchive clears it, kill's deleted_at is untouched (so BE-1 revive is unaffected). The list hides archived_at rows by default; a new include_archived surfaces them for an archived view — orthogonal to include_ended. Updates the root-service archive tests to the new semantics.
Adds an archive action to the session rail and history popover, backed by the new archived_at server state. The durable list query now includes archived rows so the reconciler carries an `archived` flag rather than mistaking an archived session for a hard-delete and pruning it; the flag hides the row from the main history/tabs and surfaces it under a collapsible "Archived" section with restore + delete. Archive/unarchive fire the server archiveSession/unarchiveSession for server-known sessions so the state syncs across devices, and stay purely local otherwise.
Regenerated the TS client from the #5436 backend OpenAPI spec so `SessionQueryRequest` carries `include_ended`/`include_archived` and `SessionStream` carries `archived_at` as first-class typed fields, and dropped the runtime cast on `querySessions` that stood in for them. Pulls in the rest of the #5436 sessions/mounts drift the checked-in client predated (SessionTurn `turn_id`, a SessionTurnComplete request, the mount file-order enum). No new tsc error signatures introduced (verified against the pre-regen baseline).
_start_turn only cleared deleted_at (the killed-tombstone revive), so starting a turn on an archived-but-not-killed session updated its flags to alive/running while archived_at stayed set — a live session absent from the default list. Now a live row that carries archived_at is un-hidden on new activity (clears only when actually archived, so the hot path is untouched for the common case).
The durable list filters archived_at per project (the default list excludes it, the archived view selects it), so back that predicate with an index instead of leaning on the created_at index alone. Mirrors ix_session_streams_project_id_created_at; migration 019.
A session started in another tab/device showed as "Untitled chat" in the durable list until opened, because the list label comes only from the server header `name` (set on explicit rename) and an adopted session has no local messages to derive a label from. Now the first user message auto-titles the session (capped, synced to the durable header via setSessionHeader), so the list is labeled everywhere before it's opened. The write atom no-ops once the session already has a title, so it fires at most once and never overwrites an explicit rename.
ardaerzin
marked this pull request as draft
July 24, 2026 16:14
ardaerzin
marked this pull request as ready for review
July 24, 2026 16:15
dosubot
Bot
added
the
size:XL
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
📝 WalkthroughWalkthroughThe PR adds a persistent ChangesSession archiving
Estimated code review effort: 4 (Complex) | ~60 minutes Sequence Diagram(s)sequenceDiagram
participant AgentChatPanel
participant useReconcileServerSessions
participant querySessions
participant SessionHistoryMenu
AgentChatPanel->>useReconcileServerSessions: reconcile scope
useReconcileServerSessions->>querySessions: fetch project sessions
querySessions-->>useReconcileServerSessions: validated session summaries
useReconcileServerSessions->>SessionHistoryMenu: update local session state
SessionHistoryMenu->>querySessions: archive or unarchive session
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
Author
|
@coderabbitai review |
✅ Action performedReview finished.
|
ardaerzin
added
Frontend
Backend
size:XL
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
api/oss/src/core/sessions/streams/service.py (1)
497-511: 🩺 Stability & Availability | 🟠 Major | 🏗️ Heavy liftSerialize archive and turn-start transitions.
A concurrent archive can commit after the flag update but before the stale
updated.archived_atcheck, leaving a running session hidden from the default list. A concurrent hard delete can likewise leave the revive retry empty while this method still returns a turn ID. Make archive/start/revive one per-session atomic transition (or use a shared lock), and fail or retry unless the final row update succeeds.Also applies to: 580-612
🧹 Nitpick comments (1)
web/oss/src/components/AgentChatSlice/state/sessions.ts (1)
1-13: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick winRemove or wire
routerAppIdAtominto the scoped session mutations.
routerAppIdAtomis unused after the import, whilesetSessionHeader/archive/unarchive/delete calls here, pluskillSessioncallers, only passsessionId/projectId. If these operations are meant to be app-scoped, use the current scope atom in these remote calls instead of leaving it unused here.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 758abb71-75b2-4b7f-8a81-46e68c99c81f
⛔ Files ignored due to path filters (15)
web/packages/agenta-api-client/src/generated/api/resources/index.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/mounts/client/Client.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/mounts/client/requests/GetMountFilesRequest.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/mounts/index.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/mounts/types/GetMountFilesRequestOrder.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/mounts/types/index.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/sessions/client/Client.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/sessions/client/requests/SessionQueryRequest.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/sessions/client/requests/SessionTurnAppendRequest.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/sessions/client/requests/SessionTurnCompleteRequest.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/resources/sessions/client/requests/index.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/types/MountFile.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/types/SessionStream.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/types/SessionTurn.tsis excluded by!**/generated/**web/packages/agenta-api-client/src/generated/api/types/index.tsis excluded by!**/generated/**
📒 Files selected for processing (23)
api/oss/databases/postgres/migrations/core_oss/versions/oss000000018_add_session_streams_archived_at.pyapi/oss/databases/postgres/migrations/core_oss/versions/oss000000019_index_session_streams_archived_at.pyapi/oss/src/apis/fastapi/sessions/models.pyapi/oss/src/apis/fastapi/sessions/router.pyapi/oss/src/core/sessions/dtos.pyapi/oss/src/core/sessions/service.pyapi/oss/src/core/sessions/streams/dtos.pyapi/oss/src/core/sessions/streams/interfaces.pyapi/oss/src/core/sessions/streams/service.pyapi/oss/src/dbs/postgres/sessions/streams/dao.pyapi/oss/src/dbs/postgres/sessions/streams/dbes.pyapi/oss/src/dbs/postgres/sessions/streams/mappings.pyapi/oss/tests/pytest/unit/sessions/test_resume_after_kill_revives_tombstone.pyapi/oss/tests/pytest/unit/sessions/test_sessions_root_service.pyweb/oss/src/components/AgentChatSlice/AgentChatPanel.tsxweb/oss/src/components/AgentChatSlice/AgentConversation.tsxweb/oss/src/components/AgentChatSlice/components/SessionHistoryMenu.tsxweb/oss/src/components/AgentChatSlice/components/SessionRail.tsxweb/oss/src/components/AgentChatSlice/state/projectSessions.tsweb/oss/src/components/AgentChatSlice/state/sessions.tsweb/packages/agenta-entities/src/session/api/api.tsweb/packages/agenta-entities/src/session/core/schema.tsweb/packages/agenta-entities/src/session/index.ts
Comment on lines
+1
to
+9
| """A resume after a STOP_KILLS_SESSION kill (or archive) must re-nest the durable row. | ||
|
|
||
| Kill/archive soft-deletes the `session_streams` row (`deleted_at`). The unique constraint on | ||
| `(project_id, session_id)` is full, so the tombstone keeps occupying the slot: on the next turn | ||
| `_start_turn` reads `None` (get filters `deleted_at IS NULL`), `create` hits the unique slot -> | ||
| `SessionStreamAlreadyExists`, and the follow-up `update` (also `deleted_at IS NULL`) matches | ||
| nothing -> no-op. Without a revive, Redis goes alive but the durable row stays a dead tombstone, | ||
| so the session vanishes from the list and `fetch`/`heartbeat` return `None`. `_start_turn` must | ||
| detect the no-match update and revive that same row (clear `deleted_at`) so resume re-nests it. |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win
Correct the archive lifecycle description.
The docstring still says archive soft-deletes the row via deleted_at, but this change makes archive preserve deleted_at and set archived_at; only kill creates the tombstone.
Proposed wording update
-"""A resume after a STOP_KILLS_SESSION kill (or archive) must re-nest the durable row.
+"""A resume after a STOP_KILLS_SESSION kill must re-nest the durable row.
-Kill/archive soft-deletes the `session_streams` row (`deleted_at`).
+Kill soft-deletes the `session_streams` row (`deleted_at`); archive keeps it live
+and uses `archived_at` to hide it from the default list.📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| """A resume after a STOP_KILLS_SESSION kill (or archive) must re-nest the durable row. | |
| Kill/archive soft-deletes the `session_streams` row (`deleted_at`). The unique constraint on | |
| `(project_id, session_id)` is full, so the tombstone keeps occupying the slot: on the next turn | |
| `_start_turn` reads `None` (get filters `deleted_at IS NULL`), `create` hits the unique slot -> | |
| `SessionStreamAlreadyExists`, and the follow-up `update` (also `deleted_at IS NULL`) matches | |
| nothing -> no-op. Without a revive, Redis goes alive but the durable row stays a dead tombstone, | |
| so the session vanishes from the list and `fetch`/`heartbeat` return `None`. `_start_turn` must | |
| detect the no-match update and revive that same row (clear `deleted_at`) so resume re-nests it. | |
| """A resume after a STOP_KILLS_SESSION kill must re-nest the durable row. | |
| Kill soft-deletes the `session_streams` row (`deleted_at`); archive keeps it live | |
| and uses `archived_at` to hide it from the default list. The unique constraint on | |
| `(project_id, session_id)` is full, so the tombstone keeps occupying the slot: on the next turn | |
| `_start_turn` reads `None` (get filters `deleted_at IS NULL`), `create` hits the unique slot -> | |
| `SessionStreamAlreadyExists`, and the follow-up `update` (also `deleted_at IS NULL`) matches | |
| nothing -> no-op. Without a revive, Redis goes alive but the durable row stays a dead tombstone, | |
| so the session vanishes from the list and `fetch`/`heartbeat` return `None`. `_start_turn` must | |
| detect the no-match update and revive that same row (clear `deleted_at`) so resume re-nests it. | |
| """ |
Comment on lines
87
to
94
| async def archive(self, *, project_id, user_id, session_id): | ||
| self.archive_calls.append( | ||
| {"project_id": project_id, "user_id": user_id, "session_id": session_id} | ||
| ) | ||
| if self.row is not None: | ||
| self.row = self.row.model_copy( | ||
| update={"deleted_at": "2026-01-01T00:00:00Z"} | ||
| update={"archived_at": "2026-01-01T00:00:00Z"} | ||
| ) |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "== locate relevant files =="
git ls-files | rg 'api/oss/tests/pytest/unit/sessions/test_sessions_root_service.py|api/oss/src/.*SessionStream|src|sessions' | head -200
echo
echo "== inspect target test =="
sed -n '1,160p' api/oss/tests/pytest/unit/sessions/test_sessions_root_service.py | cat -n
echo
echo "== find SessionStream/session model definitions =="
rg -n "class SessionStream|archived_at|SessionStream" api/oss/src api/oss/tests -g '*.py' | head -200Repository: Agenta-AI/agenta
Length of output: 8994
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "== deterministic model_copy behavior probe =="
python3 - <<'PY'
import sys
try:
from datetime import datetime, timezone
from pydantic import BaseModel
except Exception as e:
print(f"import_error: {type(e).__name__}: {e}")
sys.exit(0)
class Model(BaseModel):
archived_at: datetime
for value in [None, "2026-01-01T00:00:00Z", datetime(2026, 1, 1, tzinfo=timezone.utc)]:
m = Model(archived_at=None, row=None)
print("- value", repr(value), "type", type(value).__name__)
try:
m2 = m.model_copy(update={"archived_at": value, "row": m.row})
print(" copied archived_at:", repr(m2.archived_at), "type:", type(m2.archived_at).__name__)
except Exception as e:
print(" copy_error:", type(e).__name__ + ":", str(e))
# Compare to construction validation.
for value in [None, "2026-01-01T00:00:00Z", datetime(2026, 1, 1, tzinfo=timezone.utc)]:
try:
m = Model(archived_at=value, row=None)
print(" constructed archived_at:", repr(m.archived_at), "type:", type(m.archived_at).__name__)
except Exception as e:
print(" construct_error:", type(e).__name__ + ":", str(e))
PYRepository: Agenta-AI/agenta
Length of output: 810
🏁 Script executed:
#!/bin/bash
set -u
echo "== python/pydantic availability =="
python3 - <<'PY'
from datetime import datetime, timezone
try:
import pydantic
print("pydantic", pydantic.__version__)
from pydantic import BaseModel
print("ok pydantic imported")
except Exception as e:
print("import_error", type(e).__name__ + ":", str(e))
PY
python3 - <<'PY'
from datetime import datetime, timezone
import pydantic
print("pydantic", pydantic.__version__)
from pydantic import BaseModel
class Model(BaseModel):
archived_at: datetime | None = None
row: str | None = None
for value in [None, "2026-01-01T00:00:00Z", datetime(2026, 1, 1, tzinfo=timezone.utc)]:
m = Model()
try:
m = m.model_copy(update={"archived_at": value})
print("input", repr(value), "output", repr(m.archived_at), "type", type(m.archived_at).__name__)
except Exception as e:
print("input", repr(value), "copy_error", type(e).__name__ + ":", str(e))
PY
echo
echo "== focused test file =="
if [ -f api/oss/tests/pytest/unit/sessions/test_sessions_root_service.py ]; then
sed -n '1,180p' api/oss/tests/pytest/unit/sessions/test_sessions_root_service.py | cat -n
else
echo "missing target file"
fi
echo
echo "== focused SessionStream definitions/usages =="
rg -n "class SessionStream|archived_at|SessionStream" api/oss/src api/oss/tests || trueRepository: Agenta-AI/agenta
Length of output: 46919
Keep the archive mock’s timestamp type-faithful.
SessionStream.archived_at is modeled as Optional[datetime], and model_copy(update=...) in the supported Pydantic version preserves unvalidated strings. Use a timezone-aware datetime here so the fake row matches production’s typed value.
Proposed fix
- update={"archived_at": "2026-01-01T00:00:00Z"}
+ update={"archived_at": datetime(2026, 1, 1, tzinfo=timezone.utc)}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| async def archive(self, *, project_id, user_id, session_id): | |
| self.archive_calls.append( | |
| {"project_id": project_id, "user_id": user_id, "session_id": session_id} | |
| ) | |
| if self.row is not None: | |
| self.row = self.row.model_copy( | |
| update={"deleted_at": "2026-01-01T00:00:00Z"} | |
| update={"archived_at": "2026-01-01T00:00:00Z"} | |
| ) | |
| async def archive(self, *, project_id, user_id, session_id): | |
| self.archive_calls.append( | |
| {"project_id": project_id, "user_id": user_id, "session_id": session_id} | |
| ) | |
| if self.row is not None: | |
| self.row = self.row.model_copy( | |
| update={"archived_at": datetime(2026, 1, 1, tzinfo=timezone.utc)} | |
| ) |
The unit fakes stamped datetime lifecycle fields with ISO strings via model_copy(update=...), which does not re-validate, so archived_at/deleted_at stayed str where the DTOs declare Optional[datetime]. Use real datetimes throughout. Also correct two stale docstrings that described archive as soft-deleting via deleted_at; archive round-trips archived_at and never touches the kill tombstone path.
Contributor
Railway Preview Environment
Updated at 2026-07-28T07:12:47.529Z |
ashrafchowdury
approved these changes
Jul 27, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Context
The agent-chat session list was localStorage-only. It lived in a single browser, so it never synced across devices or tabs, and a delete or rename never left that browser. Open the app elsewhere and you saw nothing; a killed session simply vanished from history. This PR makes the list server-backed and reconciled, so it is durable and consistent everywhere.
What this adds
The list is now backed by the durable
/sessions/queryendpoint and folded over the localStorage cache by a reconciler:include_endedkeeps killed-but-resumable sessions in the list with a muted "Ended" chip, and resuming a killed session revives its durable row.archived_atcolumn (distinct fromdeleted_at) gives a hide-but-recoverable state, with archive/unarchive actions in the rail and history popover, an "Archived" section, andinclude_archivedfor the archived view. A live turn on an archived session auto-unarchives it.Before: the session list was one browser's localStorage; delete, rename, and kill stayed local; there was no archive.
After: the list is durable server streams reconciled over localStorage; delete, rename, archive, and ended state sync across devices.
Backend
session_streamsgainsarchived_at(migrationoss018) plus an index on(project_id, archived_at)(oss019).archive/unarchiveusearchived_atso kill'sdeleted_atpath is untouched.include_endedandinclude_archivedare threaded through the query path and filter independently.include_ended,include_archived, andarchived_atare typed rather than cast at runtime.Tests / notes
oss018revisesoss017(the current base head).generate.sh --localbefore merge guarantees the client matches the backend.What to QA