chutes-build 2.0.4 → 2.0.6
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +62 -0
- package/package.json +7 -7
package/CHANGELOG.md
CHANGED
|
@@ -4,6 +4,68 @@ All notable changes to Chutes Build will be documented in this file. The format
|
|
|
4
4
|
is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this
|
|
5
5
|
project follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
|
|
6
6
|
|
|
7
|
+
## [2.0.6] - 2026-10-26
|
|
8
|
+
|
|
9
|
+
### Added
|
|
10
|
+
|
|
11
|
+
- `chutes-build routing` and the `/routing` slash command show the saved
|
|
12
|
+
Chutes model-routing pool, the three strategies, and live TTFT/TPS from
|
|
13
|
+
`GET /invocations/stats/llm`. Pool members and stats are resolved by
|
|
14
|
+
`chute_id` against the live catalogue, so the view shows model names rather
|
|
15
|
+
than private chute labels.
|
|
16
|
+
- The routing view is read-only. Alias writes are not a supported surface yet:
|
|
17
|
+
the Chutes endpoint is eventually consistent and intermittently answers 502,
|
|
18
|
+
so the write path stays hidden until the platform stabilises.
|
|
19
|
+
- When the account has a saved `default` routing pool, Auto targets
|
|
20
|
+
`default:latency` instead of composing an inline pool, so the server-side
|
|
21
|
+
routing choice made in the Chutes web app is honoured.
|
|
22
|
+
- Reasoning-effort controls are exposed only for models whose accepted values
|
|
23
|
+
were probed against the Chutes endpoint. The catalogue publishes the
|
|
24
|
+
`reasoning` feature but not the effort menu, so the verified value set is used
|
|
25
|
+
instead of the generic fallback.
|
|
26
|
+
- The external-provider `deepseek-ai/DeepSeek-V4-Pro-0813` and
|
|
27
|
+
`Qwen/Qwen3.8-Max-0902` rows carry no feature list, so their effort menus come
|
|
28
|
+
from endpoint-verified values (`low|medium|high` and `low|medium|xhigh`).
|
|
29
|
+
`DeepSeek-V4-Flash` and `Kimi-K3` use their verified value sets too. The
|
|
30
|
+
2026-10-26 beta wave adds `DeepSeek-V4.1-Flash` (`low|medium|high`), `GLM-5.3`
|
|
31
|
+
and `GLM-5.2-Fast` (`low|high|max`), `Kimi-K2.7-Code`
|
|
32
|
+
(`low|medium|high|max`), and `Qwen3.8-2.4T-A95B` (`low|medium|xhigh`). All of
|
|
33
|
+
these are PAYG-only non-TEE rows; Auto still leaves them out, so selecting
|
|
34
|
+
them is explicit and never spends PAYG silently.
|
|
35
|
+
|
|
36
|
+
### Fixed
|
|
37
|
+
|
|
38
|
+
- The live invocation-stats body is tens of megabytes: fetch it with a longer
|
|
39
|
+
deadline and `Accept-Encoding: identity` so the routing view can decode it
|
|
40
|
+
behind edges that negotiate an unsupported compression.
|
|
41
|
+
- `/routing` does not block the TUI on the stats fetch: a cold or stale cache
|
|
42
|
+
starts a background refresh and asks the user to run the command again.
|
|
43
|
+
- Auto excludes `confidential_compute: false` models explicitly instead of
|
|
44
|
+
relying on their missing feature list, so a PAYG model can never join the
|
|
45
|
+
automatic pool.
|
|
46
|
+
- The saved-pool alias cache is seeded on every interactive startup path, not
|
|
47
|
+
only the headless one, so Auto can honour `default:latency` in the TUI and
|
|
48
|
+
leader modes.
|
|
49
|
+
- Credential headers are marked sensitive, so keys cannot appear in HTTP debug
|
|
50
|
+
logs even for management routes that need the raw-key form.
|
|
51
|
+
- The zstd stack is on the patched 0.14 line (RUSTSEC-2026-0391, -0398,
|
|
52
|
+
-0403, -0410) and the unused tonic zstd feature is gone; `cargo deny` is
|
|
53
|
+
clean again.
|
|
54
|
+
|
|
55
|
+
## [2.0.5] - 2026-10-25
|
|
56
|
+
|
|
57
|
+
### Fixed
|
|
58
|
+
|
|
59
|
+
- **External-provider 402s no longer masquerade as the weekly limit.** New
|
|
60
|
+
non-TEE catalog entries such as `deepseek-ai/DeepSeek-V4-Pro-0813` and
|
|
61
|
+
`Qwen/Qwen3.8-Max-0902` run on external providers and require a positive PAYG
|
|
62
|
+
balance; Chutes answers `402 A positive balance is required for external
|
|
63
|
+
provider execution`. The pager treated every 402 as a credit-limit denial and
|
|
64
|
+
opened "You hit your weekly limit" even when the subscription window was
|
|
65
|
+
nearly empty. Now that specific 402 is shown with the provider's message, and
|
|
66
|
+
non-TEE models carry a `PAYG` warning notice in the model picker. The
|
|
67
|
+
subscription/weekly quota is not involved.
|
|
68
|
+
|
|
7
69
|
## [2.0.4] - 2026-10-25
|
|
8
70
|
|
|
9
71
|
### Added
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "chutes-build",
|
|
3
|
-
"version": "2.0.
|
|
3
|
+
"version": "2.0.6",
|
|
4
4
|
"description": "Privacy-first, open-source coding agent for the Chutes ecosystem",
|
|
5
5
|
"license": "Apache-2.0",
|
|
6
6
|
"author": { "name": "Michael Gasperini", "email": "michael@mikesoft.it", "url": "https://mikesoft.it" },
|
|
@@ -37,12 +37,12 @@
|
|
|
37
37
|
"check": "npm test && npm run verify:release && npm pack --dry-run"
|
|
38
38
|
},
|
|
39
39
|
"optionalDependencies": {
|
|
40
|
-
"chutes-build-darwin-arm64": "2.0.
|
|
41
|
-
"chutes-build-darwin-x64": "2.0.
|
|
42
|
-
"chutes-build-linux-arm64-gnu": "2.0.
|
|
43
|
-
"chutes-build-linux-x64-gnu": "2.0.
|
|
44
|
-
"chutes-build-win32-arm64": "2.0.
|
|
45
|
-
"chutes-build-win32-x64": "2.0.
|
|
40
|
+
"chutes-build-darwin-arm64": "2.0.6",
|
|
41
|
+
"chutes-build-darwin-x64": "2.0.6",
|
|
42
|
+
"chutes-build-linux-arm64-gnu": "2.0.6",
|
|
43
|
+
"chutes-build-linux-x64-gnu": "2.0.6",
|
|
44
|
+
"chutes-build-win32-arm64": "2.0.6",
|
|
45
|
+
"chutes-build-win32-x64": "2.0.6"
|
|
46
46
|
},
|
|
47
47
|
"keywords": [
|
|
48
48
|
"ai",
|