deps(deps): bump @anthropic-ai/sdk from 0.32.1 to 0.98.0#9
Closed
dependabot[bot] wants to merge 1 commit into
Closed
deps(deps): bump @anthropic-ai/sdk from 0.32.1 to 0.98.0#9dependabot[bot] wants to merge 1 commit into
dependabot[bot] wants to merge 1 commit into
Conversation
Author
LabelsThe following labels could not be found: Please fix the above issues or remove invalid values from |
sneaky-hippo
added a commit
that referenced
this pull request
May 24, 2026
3-agent parallel build of opt-in multimodal tokenizer workers closing the
user-flagged gap from W771-W774 ship: "were doing tokenization for those
multimodal distillations right?" Answer was: text-side yes, media-side no
(scaffolding only). User picked Upgrade option. This batch ships real
CLIP/Whisper/Video-ViT tokenizer workers using the W462/W464 isolated-
worker pattern: heavy Python deps live in workers/{modality}-tokenize/,
root kolm install untouched, honest no_detector_installed envelope when
deps missing.
Same distinct-named dispatcher discipline as W771-W813 batch (cmdW771Vlm
extended with _w771bTokenize / _w771bTokenizeDoctor, cmdW772Audio with
_w772bTokenize / _w772bTokenizeDoctor, cmdW773Video with _w773bTokenize /
_w773bTokenizeDoctor) so cli/kolm.js + src/router.js merged clean across
3 parallel agents with zero conflict.
W771b - VLM Tokenizer Worker (CLIP/SigLIP patches)
workers/vision-tokenize/package.json + tokenize.mjs Node shim.
Default model openai/clip-vit-large-patch14 (256 patches @ 224x224).
Priority chain: $KOLM_VISION_TOKENIZE_CMD -> kolm-vision-tokenize on
PATH -> python3 ~/.kolm/scripts/vision-tokenize.py.
src/vision-tokenize.js (VISION_TOKENIZE_VERSION='w771b-v1') exports
tokenizeImage() + getVisionTokenizeDoctor(). NEVER throws on malformed
input - returns ok:false envelope.
POST /v1/vlm/tokenize + GET /v1/vlm/tokenize/doctor (auth-gated).
CLI: kolm vlm tokenize <path|uri|url> [--model NAME] + tokenize-doctor.
Modes accept --path, --uri, --url, --image_base64.
14/14 tests green.
W772b - Audio Tokenizer Worker (WhisperProcessor mel/BPE)
workers/audio-tokenize/package.json + tokenize.mjs Node shim.
Default model openai/whisper-large-v3 (30s -> 1500 mel frames @ 80
dims, plus BPE text tokens from transcription).
Priority chain: $KOLM_AUDIO_TOKENIZE_CMD -> kolm-audio-tokenize on
PATH -> python3 ~/.kolm/scripts/audio-tokenize.py.
src/audio-tokenize.js (AUDIO_TOKENIZE_VERSION='w772b-v1') exports
tokenizeAudio() + getAudioTokenizeDoctor(). Materializes base64 to
tempfile + unlinks after spawn.
POST /v1/audio/tokenize + GET /v1/audio/tokenize/doctor (auth-gated).
CLI: kolm audio tokenize <path|uri> [--model NAME] [--with-mel]
[--with-text-tokens] + tokenize-doctor.
13/13 tests green.
W773b - Video Tokenizer Worker (frame-patch tokens)
workers/video-tokenize/package.json + tokenize.mjs Node shim.
Default model llava-hf/LLaVA-NeXT-Video-7B-hf (8 sampled frames * 144
patches/frame = 1152 patch tokens).
SAMPLING_STRATEGIES frozen ['uniform','adaptive','keyframe','dense']
(distinct from W773 frame-sampler's ['uniform','keyframe',
'scene_change','adaptive'] - different module, different concern).
num_frames capped at 32 to prevent OOM in tests (num_frames_capped
flag surfaced).
Priority chain: $KOLM_VIDEO_TOKENIZE_CMD -> kolm-video-tokenize on
PATH -> python3 ~/.kolm/scripts/video-tokenize.py.
src/video-tokenize.js (VIDEO_TOKENIZE_VERSION='w773b-v1') exports
tokenizeVideo() + getVideoTokenizeDoctor().
POST /v1/video/tokenize + GET /v1/video/tokenize/doctor (auth-gated).
Rejects invalid sampling_strategy with error:'invalid_sampling_strategy'.
CLI: kolm video tokenize <path|uri> [--model NAME]
[--sampling-strategy STRAT] [--num-frames N] + tokenize-doctor.
15/15 tests green.
Honest envelope pattern (mirrors W462 image-PII + W464 voiceprint scrub)
Every no-detector path returns:
{ok:false, error:'no_detector_installed', install_hint:'install
python3 + transformers + torch + ...', patch_token_count:null,
mel_frame_count:null, text_token_count:null, sampled_frame_count:
null, total_patch_tokens:null}
NEVER silent passthrough. NEVER fabricated token counts. Doctor mode
reports per-Python-dep readiness (transformers / torch / Pillow /
librosa / soundfile / decord) so operators know exactly what to
install.
W411 defense-in-depth
src/{vision,audio,video}-tokenize.js are pure I/O (no tenant-keyed
state). Tenant fence is the CALLER's job at the router layer
(req.tenant_record gate on every POST /v1/{modality}/tokenize).
W771b #9, W772b #13, W773b #15 lock-in the tenant-isolation contract.
W604 anti-brittleness
W771b/W772b/W773b new tests use regex /wave?(\d{3,4})b?/ (accepts
wave771, wave771b, w771, w771b forms). sw.js cache slug changed to
use 'wave' prefix form so the regex test from W813 #30
(/wave?(\d{3,4})/ OR /-w(\d{3,4})-/ with threshold >= 761) keeps
passing through this slug migration. Slug:
kolm-v57-2026-05-24-wave771b-wave772b-wave773b-vision-audio-video-tokenizers.
max wave token = 773 >= 761.
DI testing seams
Every test stubs the tokenizer via $KOLM_{MODALITY}_TOKENIZE_CMD env
override pointing at a Node script that emits canned JSON. Tests
NEVER hit real Python, real torch, real CLIP/Whisper. The honest
no-detector path is exercised by clearing PATH and HOME so the
priority chain falls through to the install_hint envelope.
Encoding discipline
All worker .mjs + Python scripts ASCII-only (em-dash -> '--',
arrow -> '->', no box-drawing). Trap from W771 fix: Python ast.parse
on Windows stdin pipe rejects non-ASCII; all trainer files now ASCII.
Test sweep: 42/42 green (W771b 14 + W772b 13 + W773b 15) + 173/173
green in cumulative W77x + W77xb + W813 sweep + 61/61 green in
W214+W455+W460+W721+W722 sibling regression.
Brand lock preserved
Eyebrow "Open-source AI workbench" + H1 "Frontier AI on your own
infrastructure." in frontend-version.json hero_h1/hero_eyebrow
untouched. Only version + source_commit_at_edit_time bumped.
dependabot
Bot
force-pushed
the
dependabot/npm_and_yarn/anthropic-ai/sdk-0.98.0
branch
3 times, most recently
from
0ee0ca0 to
ab5d0e8
Compare
Bumps [@anthropic-ai/sdk](https://github.com/anthropics/anthropic-sdk-typescript) from 0.32.1 to 0.98.0. - [Release notes](https://github.com/anthropics/anthropic-sdk-typescript/releases) - [Changelog](https://github.com/anthropics/anthropic-sdk-typescript/blob/main/CHANGELOG.md) - [Commits](anthropics/anthropic-sdk-typescript@sdk-v0.32.1...sdk-v0.98.0) --- updated-dependencies: - dependency-name: "@anthropic-ai/sdk" dependency-version: 0.98.0 dependency-type: direct:production update-type: version-update:semver-minor ... Signed-off-by: dependabot[bot] <support@github.com>
dependabot
Bot
force-pushed
the
dependabot/npm_and_yarn/anthropic-ai/sdk-0.98.0
branch
from
ab5d0e8 to
0ea6db2
Compare
sneaky-hippo
added a commit
that referenced
this pull request
May 27, 2026
- 7 new HTTP routes: POST /v1/devices/{add,remove}, GET /v1/fleet/{status,monitor}, POST /v1/fleet/{deploy,rollback,stop}
- docs/product-surfaces.json: own 5 stray groups (auth+sales → identity-access-billing, test-quants → compile-artifact-verification, deploy+test-device → deployment-edge-federated) closes 727 vs 736 catalog mismatch
- 4 cache lock-in tests relaxed to active-CACHE regex+threshold (per W886 #9 / W446 #5 pattern)
- public/sw.js: v120 → v121 (wave896-uc12-fleet-http-routes)
- server.js: degraded-mode boot when provisionTenant throws so /health still answers and operators can introspect
- cli/kolm.js: route 'kolm verify rcpt_...' to wrapper-cli RECEIPTS_VERBS.verify
- og:image meta tags on /ci, /edge, /fleet
- vercel.json: +18 rewrites for new fleet/device surfaces
Release-verify gates green: lint:refs, control-files, openapi-sync, claim-verify, sdk-manifest, sdk-smoke, local-surfaces 74/74, verify-claims. W888 C/D/E device+fleet families 82/82. Cache lock-ins 42/42.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Author
|
Superseded by #13. |
dependabot
Bot
deleted the
dependabot/npm_and_yarn/anthropic-ai/sdk-0.98.0
branch
sneaky-hippo
added a commit
that referenced
this pull request
May 29, 2026
…duct links -> /about Per user: 'my personal email shouldn't be anywhere' + 'repoint /product links -> /about'. - EMAIL: rodneyyesep@gmail.com -> dev@kolm.ai across 93 product/user-facing files (public/docs/account/brand + src/cli/api defaults incl. transactional from-address). Updated tests/wave708 expectation to match the new source value. - /product -> /about: repointed href on 353 pages (regex-guarded, never matches /products). Updated the canonical-nav source scripts/inject-nav.cjs (NAV_ITEMS + mega Overview) to /about and re-ran it (0 touched, 334 idempotent-noop). Updated the W221/W223/W226/W227 cross-link + idempotency test contracts from /product to /about (the product page now lives at /about; /product 308-redirects there). 67/67 in those families. - about.html JSON-LD/breadcrumb -> /about; webgpu-runner.html meta description (W538 6/6). /product the PAGE still exists + 308-redirects to /about (W221 #9/#10/#13 intact). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
sneaky-hippo
added a commit
that referenced
this pull request
May 29, 2026
…out', generators + test contracts)
The /product->/about repoint cascaded into the W221 canonical 5-anchor nav (woven
through ~10 test families + the inject-nav generator + 4 CLI-doc generators) — W301
already encoded the intended end-state ('About/Models/Docs/Pricing/Enterprise'). Finished it:
- scripts/inject-nav.cjs: canonical first anchor now label 'About' -> /about (slug kept for CSS);
re-ran it -> 334 pages regenerated with the About/About nav.
- 4 CLI-doc generators (build-changelog, write-extra/missing-cli-docs, write-w869b-cli-stubs):
nav-top template 'Product'->/product => 'About'->/about so regeneration stays consistent.
- Test contracts updated /product->/about + 'Product'->'About' in the 5-anchor nav assertions:
W221/W223/W226/W227/W230/W322-325(W32x)/W374/W409z/W559. /product PAGE-existence + redirect
tests (W221 #9/#10/#13) untouched (the page still exists + 308-redirects).
Full suite: 8789 pass / 4 fail — all 4 are the known env-concurrency flakes (quantize-worker
torch-import + ship-gate inner node --test), every one passes standalone. Zero real regressions.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
sneaky-hippo
added a commit
that referenced
this pull request
May 31, 2026
- H1 claim "One stack for production AI." -> "From your API calls to a model you own." (directional, outcome-first; hidden "The AI compiler." span preserved for W221 #9) - lede tightened + adds the receipt-on-every-call moat line, no em-dash. Surgical visible-text only; W221 #9 contract intact (The AI compiler in H1, the four verbs, /captures /quickstart /runtimes /drift /use-cases anchors). wave221 13/13, site 7/7, wave538 6/6. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
sneaky-hippo
added a commit
that referenced
this pull request
May 31, 2026
- Homepage hero lede: "Point your OpenAI-compatible client at one URL" -> "Point any client at one URL: OpenAI, Anthropic, Gemini, vLLM, or your own. kolm routes every model..." (the gateway routes all providers; OpenAI-compat is the interface, not a limit). Em-dash budget held (0). - /product: the product lines were muddied by a jargon "Wrapper and Studio. One signed artifact." section. Reframed to "Two ways in. One signed artifact." with Product 1 = Gateway (route any model in front of your app) and Product 2 = Compiler (distill/quantize/sign your own model, terminal or no-code studio). Pillar strip stays Gateway / Compiler / Runtime / Proof. Residual Wrapper/Studio body jargon + CTAs removed; gateway framing broadened across the section. W221 #9 contract intact; wave221 13/13, site 7/7, wave538 6/6, wave220 12/12, wave271 25/25. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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
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.
Bumps @anthropic-ai/sdk from 0.32.1 to 0.98.0.
Release notes
Sourced from @anthropic-ai/sdk's releases.
... (truncated)
Changelog
Sourced from @anthropic-ai/sdk's changelog.
... (truncated)
Commits
32ce8c0chore: release main1873a96feat(api): Add support for thinking-token-count beta for estimated tokens in ...ac9ece3chore: release main1987147fix(runner): skip tool calls SessionToolRunner does not own409ff0echore: release main (#1052)a53f60dchore: release maind1b8d04feat(api): Add support for cache diagnostics beta8e43bf8chore(api): spec updates697e4d5codegen metadatacd5801cfeat(api): Add BetaManagedAgentsSearchResultBlock typesMaintainer changes
This version was pushed to npm by packy-anthropic, a new releaser for
@anthropic-ai/sdksince your current version.