chutes-build 2.0.1 → 2.0.3

package/CHANGELOG.md

@@ -4,6 +4,91 @@ All notable changes to Chutes Build will be documented in this file. The format
 is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this
 project follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.0.3] - 2026-10-24
+
+### Added
+
+- **The tool catalog is 11.5% lighter.** `browser`, `extract_document`,
+  `generate_media`, and `scheduler_create` now export a compact model-facing
+  schema: nested `description` and `examples` fields are dropped while every
+  parameter name, type, enum, required entry, and union member survives.
+  Dispatch keeps validating against the canonical schema, so behavior is
+  unchanged. The default catalog drops from 11,573 to 10,239 estimated tokens
+  across 29 tools.
+- **Latency benchmark.** `scripts/bench_chutes.py` measures TTFT, inter-token
+  latency, and total stream time against the same Chutes streaming contract
+  the CLI uses, for a model id or a routing string such as `default:latency`.
+  `docs/performance.md` explains how to compare routing strategies, read the
+  result, and account for prompt caching.
+
+### Added
+
+- **Chutes platform knowledge is available on demand.** The fixed prompt now
+  carries a compact map of the two Chutes hosts, API-key auth, the live model
+  catalog as the source of truth, and saved/inline `:latency`/`:throughput`
+  routing, and points at a new bundled `chutes-platform` skill for the full
+  platform reference: chat/vision/tool features, media chutes, OCR/PDF,
+  embeddings, quotas and billing, TEE attestation, error handling, and the
+  CLI mapping. The skill ships in the binary, is listed by the `skill` tool,
+  and can be shadowed by a user or repository skill.
+- **Independent tool calls batch in one response.** The prompt now tells the
+  model to issue non-dependent calls together, which the parallel dispatch
+  path already runs concurrently, cutting a round trip on multi-tool work.
+
+### Removed
+
+- **The dormant legacy Imagine media tools are gone.** `image_gen`,
+  `image_edit`, `image_to_video`, and `reference_to_video` targeted the xAI
+  Imagine API, which Chutes does not serve; `ImageGenTool`, `ImageEditTool`,
+  `ImageToVideoTool`, `ReferenceToVideoTool`, `ImageGenClient`,
+  `VideoGenClient`, their config plumbing, and the upstream Imagine model ids
+  and legacy video storage prefix are deleted. `/imagine` and
+  `/imagine-video` are unchanged and use the Chutes-native `generate_media`
+  tool. The related configuration keys (`features.image_gen`,
+  `features.video_gen`, the model overrides, `zdr_video_output_s3`, and the
+  `imagine_tools_disabled` remote setting) are still parsed so existing
+  configuration files keep loading, but they are inert.
+
+## [2.0.2] - 2026-10-23
+
+### Fixed
+
+- **The palette's feedback entry opens GitHub.** "Send Feedback" still launched
+  the dormant in-app feedback modal, which made the spec's "unreachable" claim
+  false and advertised a flow the build cannot complete. The entry now runs
+  `/feedback`, so it opens the public issue form like the command does.
+- **`/feedback <text>` keeps the report.** GitHub issue forms ignore a URL
+  body prefill, so the command copies the message to the clipboard before
+  opening the issue-form chooser.
+- **Web search and `deep-research` work again.** The Chutes router serves Chat
+  Completions, not the OpenAI Responses API, so the old client hit
+  `POST /v1/responses` and a 404 HTML page. With a Chutes base URL the
+  `web_search` tool now queries the direct public search provider
+  (DuckDuckGo's HTML endpoint), sends no Chutes credential to it, returns
+  titles/URLs/snippets with the configured domain filters applied, and leaves
+  model-hosted search intact for custom Responses-capable endpoints.
+- **A build without a feedback backend no longer advertises `send_feedback`.**
+  The tool is gated on the telemetry capability the Chutes distribution withholds,
+  so its schema no longer costs ~1.2k prompt tokens on every turn while every call
+  would be refused; the default catalog is 29 tools / 11,573 estimated tokens
+  (was 30 / 12,784).
+- **Media generation follows the live catalog.** `/imagine` and
+  `/imagine-video` no longer pin a model name: a bare command lists the
+  currently available image/video models, `<description>` lets the tool pick
+  an available model, and `<model> <description>` (or `--model`) selects one
+  explicitly. `generate_media` resolves catalog entries by Chutes chute id
+  (the account `describe` endpoint 404s on slugs), classifies video before
+  image so image-to-video chutes never serve image requests, and falls back
+  only to available generation models of the requested kind.
+- **Clearing the API key refreshes the model catalog.** The catalog credential
+  refresh now keys off any credential change, so removing a key cannot leave
+  the catalogue on the stale `ApiKey` identity.
+
+### Removed
+
+- The obsolete PTY tests that drove the in-app feedback modal and its drafts;
+  the plan-overlay command test now asserts the opened GitHub URL instead.
+
 ## [2.0.1] - 2026-10-23
 
 ### Fixed

package/PRIVACY.md

@@ -68,7 +68,7 @@ explicit hosted feature:
 | `api.chutes.ai` | Chutes account and media APIs | account requests, media parameters and user-selected media assets; Chutes API credential |
 | `chutes.ai/docs`, `chutes.ai/news` | mandatory primary-source verification for Chutes topics | requested official page URLs; no Chutes API credential |
 | `context7.com` | current library documentation | library names and documentation queries; known secrets are rejected and an optional Context7 key is sent only to the official HTTPS service |
-| DuckDuckGo or Brave Search | web search | search query; Brave receives only its dedicated key |
+| `html.duckduckgo.com` | web search | search query; no Chutes API credential |
 | User-selected web pages | web fetch/browser automation | normal HTTP/browser traffic required by the requested action |
 | User-configured MCP servers/plugins | external tools | data required by the selected tool and its own configuration |
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "chutes-build",
-  "version": "2.0.1",
+  "version": "2.0.3",
   "description": "Privacy-first, open-source coding agent for the Chutes ecosystem",
   "license": "Apache-2.0",
   "author": { "name": "Michael Gasperini", "email": "michael@mikesoft.it", "url": "https://mikesoft.it" },
@@ -37,12 +37,12 @@
     "check": "npm test && npm run verify:release && npm pack --dry-run"
   },
   "optionalDependencies": {
-    "chutes-build-darwin-arm64": "2.0.1",
-    "chutes-build-darwin-x64": "2.0.1",
-    "chutes-build-linux-arm64-gnu": "2.0.1",
-    "chutes-build-linux-x64-gnu": "2.0.1",
-    "chutes-build-win32-arm64": "2.0.1",
-    "chutes-build-win32-x64": "2.0.1"
+    "chutes-build-darwin-arm64": "2.0.3",
+    "chutes-build-darwin-x64": "2.0.3",
+    "chutes-build-linux-arm64-gnu": "2.0.3",
+    "chutes-build-linux-x64-gnu": "2.0.3",
+    "chutes-build-win32-arm64": "2.0.3",
+    "chutes-build-win32-x64": "2.0.3"
   },
   "keywords": [
     "ai",