Aliou Diallo

pi-neuralwatt releases

November 2026

pi-neuralwatt: v0.21.0

Minor Changes ebddddb: Add /neuralwatt:flex to toggle between standard and flex model variants, including models routed through a virtual model. Patch Changes ed3cf03: Seed System One classifier models in the offline fallback table and verify them in the model drift check, shari…

pi-neuralwatt: v0.20.0

Minor Changes ab5888b: Track daily free energy from the new daily_free_energy block on /v1/quota: a Daily Free Energy bar in the Subscription tab with a per-billing-period total, a quota-warning stage when the day's allowance runs low (re-arms at UTC rollover, configurable via q…

pi-neuralwatt: v0.19.3

Patch Changes 1f5106f: Sync the offline fallback model table with the live Neuralwatt API: remove the retired Kimi K3 models (kimi-k3, kimi-k3-fast, kimi-k3-flex) and update the nw-large tracking alias to its new target, Kimi K3.1 Pro (lower pricing and the K3.1 Pro effort contr…

pi-neuralwatt: v0.19.2

Patch Changes 0c42dbd: Add Kimi K3.1 Pro and Kimi K3.1 Pro (flex) to the offline fallback model table (fixes #146). New Moonshot flagship family: 1M context, vision, reasoning efforts max/high/medium/low/none with xhigh → max and minimal → low aliases; flex variant priced at the…

pi-neuralwatt: v0.19.0

Minor Changes d5b189c: Add a hostedTools.enabled setting that, when explicitly set to false, sends the X-NW-Tools-Opt-Out header on OpenAI-surface requests to disable Neuralwatt hosted tools. Unset leaves requests untouched. f029a14: Add a standalone @aliou/pi-neuralwatt/provide…

pi-neuralwatt: v0.18.1

Patch Changes feb140a: Remove DeepSeek V4 Flash family from the offline fallback models. The model and its -flex and -speed variants are no longer served by the Neuralwatt API; DeepSeek V4.1 Flash remains available. 883818b: Apply project settings only after Pi trusts the projec…

October 2026

pi-neuralwatt: v0.18.0

Minor Changes f5a2e13: Remove the built-in sub-bar integration. The subBarIntegration feature (live quota usage in the status bar) is no longer part of the package; it ships as a standalone one-off pi extension distributed as a GitHub gist (https://gist.github.com/aliou/9939544f…

pi-neuralwatt: v0.17.0

Minor Changes 53c4e26: Add provider.apiBaseUrl config. It sets the base URL for the extension's catalog and quota requests and is the fallback for models that carry no baseUrl of their own — models.json overrides, custom providers, and proxy routing all take precedence. Defaults…

September 2026

pi-neuralwatt: v0.16.2

Patch Changes 4bcbe4c: Fix reasoning replay on the openai-completions surface: prior-turn chain-of-thought is replayed in the reasoning field (the field K3 streams), but the served chat templates only render reasoning_content, so the model never saw its earlier thinking on any t…

pi-neuralwatt: v0.16.1

Patch Changes 2dcc7ce: Model sync: add DeepSeek V4.1 Flash (+ flex) with its full reasoning ladder, raise DeepSeek V4 Flash's max output tokens to 393216, and snapshot effort_aliases for every reasoning family so the anthropic-messages thinking map resolves unsupported levels of…

pi-neuralwatt: v0.16.0

Minor Changes e7b6b5d: Add a provider.api setting (/neuralwatt:settings → API, default openai-completions) that serves the catalog on Neuralwatt's Anthropic-compatible POST /v1/messages endpoint (vLLM-backed) — native tool use and thinking streams with the same model ids. Applie…

pi-neuralwatt: v0.15.4

Patch Changes 45373d6: Take flex model pricing from the API's already-discounted values instead of applying a local 0.65 multiplier at catalog build time. 48b331e: Sync the offline fallback model table with the live Neuralwatt API: remove GLM-5.2 and its variants (gone upstream)…

pi-neuralwatt: v0.15.3

Patch Changes 51f9dbf: fix: exclude embeddings model from catalog and harden resolveMaxTokens d15f9c8: Make the model-store cache scaffold-aware. A fresh entry persisted without a key only carries the public scope, so logged-in users never saw preview, grant-gated, or private mo…

August 2026

pi-neuralwatt: v0.15.0

Minor Changes b3caf85: Build the model catalog from the live /v1/models response instead of local visibility buckets. Model refresh now fetches anonymously when no API key is configured, builds the catalog with overrides for Flex pricing, Kimi K3 context, Qwen chat-template comp…

pi-neuralwatt: v0.13.0

Minor Changes d58a55f: Derive reasoning levels from the endpoint's supported_efforts. thinkingLevelMap is now built by identity from each model's metadata.reasoning.supported_efforts (plus mandatory) via a single buildThinkingLevelMap helper, instead of hand-tuned per-family con…

pi-neuralwatt: v0.12.0

Minor Changes fbd1b1c: Adopt @aliou/pi-utils-settings 0.19.1 versioned migrations. Bumps the dependency and adds semver version strings to all config migrations, leaving their content-based shouldRun gates intact. Switches schema generation to pi-settings-schema. 32ba87a: Add an…

July 2026
June 2026

pi-neuralwatt: v0.7.0

Minor Changes 9384557: Handle Neuralwatt stream rate limits before the OpenAI SDK drops response headers. Show layer-specific 429 messages, keep Pi auto-retry detection working, and parse SSE quota comments for live quota updates. Patch Changes 57e5ac2: Sync model list with live…

May 2026

pi-neuralwatt: v0.5.0

Minor Changes 3c467b3: Remove live model sync from provider endpoint. Models are now purely hardcoded in src/extensions/provider/models.ts and validated against the Neuralwatt /v1/models API at test time. Removed: src/lib/fetch-models.ts (live model fetch + mapApiModel) src/util…

April 2026

pi-neuralwatt: v0.2.0

Minor Changes 6f4672e: Fetch live models from Neuralwatt API on session start. The extension registers with a hardcoded model cache immediately on startup, then fetches /v1/models on session start and re-registers the provider with live data (including pricing, capabilities, and…