You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit ccf4917
Browse filesBrowse the repository at this point in the historyBrowse files
Copy file name to clipboardExpand all lines: CHANGELOG.md
+25Lines changed: 25 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -7,6 +7,31 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
7
7
8
8
## [Unreleased]
9
9
10
+
## [1.6.0] - 2026-09-23
11
+
12
+
### Added
13
+
14
+
- Channels can opt into Session presentation, per-channel Default/Comfy/Compact density, and Default/By active ordering without changing the underlying conversation model.
15
+
- Board now uses authoritative operational lanes—Working, Needs you, Scheduled, Failed, and Complete—with idle and archived sessions kept in collapsed history.
16
+
- Messages expose a direct Copy action, and ordinary Work Log steps show user-local timestamps.
17
+
- Native vision sends supported human image uploads to compatible models as image content and lets resident agents inspect workspace images through the bounded `view_image` tool.
18
+
- Clients record the user's IANA time zone so agent invocations receive both user-local time and the exact UTC instant.
19
+
20
+
### Changed
21
+
22
+
- Thread navigation is latest-intent-wins, deduplicates duplicate destinations, paints bounded cached snapshots immediately, loads channel and thread data concurrently, and server-paginates long threads while preserving the visible anchor when older replies load.
23
+
- Board and thread listings use lightweight bounded payloads so large channels avoid unnecessary database, transfer, and browser work.
24
+
- Provider cache policy is applied after the actual route is selected. Claude receives reusable stable and rolling breakpoints; ChatGPT and xAI receive stable thread-scoped cache keys; provider usage is normalized into cache reads, cache writes, uncached input, and logical input.
25
+
- Embedded ReRouted advances to 0.5.15 with bounded model validation, Claude Code 2.1.280 OAuth compatibility, cross-turn Claude cache reuse, and valid atomic serialization of parallel Claude tool results.
26
+
27
+
### Fixed
28
+
29
+
- Reader-controlled conversation and Work Log scrolling remains stable through live updates, sidebar changes, message expansion, reconnects, foreground recovery, and growing streamed replies.
30
+
- Stopping or cancelling a running scheduled wake authoritatively cancels its durable follow-up, prevents finalization from re-arming it, and keeps cancellation controls visible while it runs.
31
+
- Manual provider model tests time out cleanly instead of remaining indefinitely in a testing state.
32
+
- Photon outages no longer restart the entire 1Helm service.
33
+
- Saved channel density is authoritative on rendered session cards.
Copy file name to clipboardExpand all lines: docs/VISION.md
+11Lines changed: 11 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -41,6 +41,17 @@ sole visual home for recurring work and its chronological run history; workflow
41
41
runs do not spill into Chat, Board, or Threads, while each run remains a normal
42
42
interactive thread when opened.
43
43
44
+
## Session presentation and operational truth
45
+
46
+
Channels may opt into Session mode without changing navigation or information
47
+
architecture. The same Chat tab, sessions, order, content, labels, colors, and
48
+
thread behavior remain; only top-level Chat rows receive a compact bordered-card
49
+
presentation. Turning the mode off restores the standard Chat presentation.
50
+
51
+
Board uses operational states for every channel, regardless of Session mode.
52
+
Runtime records—not generated prose or the ambiguous database `open` value—own
53
+
Working, Needs you, Scheduled, Failed, Complete, Idle, and Archived.
54
+
44
55
## Honest model usage
45
56
46
57
A thread's primary token indicator answers how much input context the latest model call processed, not how many repeated prompt-token encounters accumulated across an agent loop. 1Helm calculates this itself from the structured messages and tool schemas it sends, using one stable provider-neutral approximation. Cached is the unchanged leading context shared with the preceding call; output counts response text and tool-call payloads received by 1Helm; calls count successful invocations. These product metrics never depend on provider usage reports. Lifetime output and model-call count remain cumulative because they represent newly generated work and actual invocations; cumulative prompt traffic is never presented as the size of a conversation.
0 commit comments