ur-agent 1.84.7 → 1.85.1

package/docs/providers.md

@@ -34,10 +34,11 @@ multimodal input, external CLI boundary, and sandbox scope:
 | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
 | Subscription | subscription | subscription-placeholder | no | no | no | n/a | n/a (no runtime) | `subscription:unconfigured` | independent subscription runtime only |
 | OpenAI API | API | UR-native | no | yes | yes | yes | UR Bash/File sandbox | `api:openai` | `OPENAI_API_KEY` |
-| Claude API | API | UR-native | no | yes | yes | yes | UR Bash/File sandbox | `api:anthropic` | `ANTHROPIC_API_KEY` |
+| Claude API | API | UR-native | no | yes | yes | yes | UR Bash/File sandbox | `api:anthropic` | `ANTHROPIC_API_KEY`; identity-linked keys also select `ANTHROPIC_WORKSPACE_ID` |
 | Gemini API | API | UR-native | no | yes | yes | yes | UR Bash/File sandbox | `api:gemini` | `GEMINI_API_KEY` |
 | OpenRouter | API/router | UR-native | no | yes | yes | yes | UR Bash/File sandbox | `api:openrouter` | `OPENROUTER_API_KEY` |
-| NVIDIA NIM | hosted/server API | UR-native | no | yes | yes | model-dependent | UR Bash/File sandbox | `api:nvidia-nim` | `NVIDIA_API_KEY`; configurable NIM endpoint |
+| NVIDIA Agentic | hosted/server API | UR-native | no | yes | yes | model-dependent | UR Bash/File sandbox | `api:nvidia-nim` | `NVIDIA_API_KEY`; public per-card endpoints or configurable NIM gateway |
+| NVIDIA Special | hosted/focused-task API | UR-native | no | task-specific | task-specific | task-specific | UR tool permission flow | `api:nvidia-special` | shared `NVIDIA_API_KEY`; exact HTTP/NVCF/gRPC contracts |
 | OpenAI-compatible | server/API | UR-native | no | yes | yes | endpoint-dependent | UR Bash/File sandbox | `openai-compatible` | optional `OPENAI_COMPATIBLE_API_KEY`; never reuses `OPENAI_API_KEY` |
 | Ollama | local/server | UR-native | no | yes | yes | yes* | UR Bash/File sandbox | `ollama` | configured local, LAN, or hosted endpoint; optional `OLLAMA_API_KEY` |
 | LM Studio | local/server | UR-native | no | yes | yes | yes | UR Bash/File sandbox | `openai-compatible:lmstudio` | configured endpoint; optional `LMSTUDIO_API_KEY` |
@@ -161,7 +162,7 @@ migrated to the previously active provider on the first provider switch or
 scoped base-URL write.
 
 The override is not limited to local runtimes. OpenAI API, Anthropic API,
-Gemini API, OpenRouter, and NVIDIA NIM can each target a separate compatible gateway using
+Gemini API, OpenRouter, and NVIDIA Agentic can each target a separate compatible gateway using
 the same command. Their official URLs are defaults, not hardcoded dispatch
 destinations; model discovery and inference use the selected provider's saved
 URL. Subscription CLI providers remain vendor-managed and do not accept a base
@@ -172,6 +173,8 @@ explicit opt-in to the native Responses adapter; it defaults to `store=false`
 and supports semantic streaming, background polling/cancellation, WebSocket
 continuation, server compaction, and deferred tool search. It does not change
 OpenAI-compatible, OpenRouter, local, or subscription-CLI providers.
+Both transports fail immediately for machine-readable permanent account or
+billing 429s; ordinary transient rate limits still honor provider retry timing.
 
 ## Provider-scoped model selection
 
@@ -212,15 +215,14 @@ or has unknown capability metadata. Arbitrary
 labels such as `deep` still require an explicit provider alias because UR
 cannot infer their rank.
 
-NVIDIA NIM is live-discovery first. UR enriches a discovered model only when
-NVIDIA's current model API reference documents that exact model's
-`reasoning_effort` values. Documented `none` appears as Minimal and `max`
-appears as Ultra while the request preserves NVIDIA's wire values. An unknown
-NIM model never inherits an invented graded ladder.
-Hosted discovery is also a positive agent-contract intersection: a row in the
-mixed NVIDIA `/v1/models` inventory is not sufficient by itself to enter the
-ongoing agent picker. Verified dedicated media/VLM endpoints are exposed only
-as one-shot task contracts and cannot pass provider/model validation.
+Public NVIDIA Agentic discovery comes from current Build Free Endpoint cards
+that advertise agent/tool use. NVIDIA's account inventory can enrich but never
+filter or remove those models. UR applies only model-specific documented
+reasoning ladders: `none` appears as Minimal and a documented `max` can map to
+Ultra while preserving NVIDIA's wire value. Unknown models never inherit a
+ladder. Focused media, analysis, embedding, safety, translation, and gRPC
+contracts are isolated in NVIDIA Special and cannot pass ongoing-agent model
+validation.
 
 For an unknown or newly released model, UR waits for provider-authored model
 metadata or a supported model-scoped probe before adding thinking parameters.
@@ -281,7 +283,18 @@ remain authoritative. See OpenRouter's
 Direct Anthropic requests retain supported prompt-cache breakpoints and enable
 per-tool `eager_input_streaming` on streaming turns. This reduces repeated
 prefill work and avoids waiting for a complete large tool argument before its
-deltas arrive. Anthropic's premium fast tier remains explicit:
+deltas arrive.
+
+Identity-linked Anthropic keys additionally need a workspace selection. UR
+accepts `provider.anthropic.workspaceId` through
+`ur config set anthropic.workspace_id wrkspc_...` or
+`ANTHROPIC_WORKSPACE_ID`, validates the `wrkspc_` resource ID, partitions the
+model cache by workspace, and sends `anthropic-workspace-id` on model
+discovery, doctor, message, streaming, and token-count requests. A provider
+400 that names this requirement is preserved with the exact configuration fix
+instead of being reduced to an unverifiable model-list failure.
+
+Anthropic's premium fast tier remains explicit:
 
 ```sh
 ur config set anthropic.speed fast
@@ -291,6 +304,8 @@ UR sends `speed: "fast"` with the `fast-mode-2026-02-01` beta only for Claude
 Opus 5 and Opus 4.8, and retains `usage.speed` so accounting can distinguish the
 tier actually served. Enabled account access is still required. Unsupported
 models stay on standard speed. See Anthropic's
+[authentication](https://platform.claude.com/docs/en/manage-claude/authentication),
+[workspace management](https://platform.claude.com/docs/en/manage-claude/workspaces),
 [prompt caching](https://platform.claude.com/docs/en/build-with-claude/prompt-caching),
 [fine-grained tool streaming](https://platform.claude.com/docs/en/agents-and-tools/tool-use/fine-grained-tool-streaming),
 and [fast mode](https://platform.claude.com/docs/en/build-with-claude/fast-mode).
@@ -298,10 +313,10 @@ and [fast mode](https://platform.claude.com/docs/en/build-with-claude/fast-mode)
 UR uses only provider-documented acceleration controls. OpenAI Responses
 already has native streaming and WebSocket continuation; Gemini 2.5+ implicit
 caching is automatic. Google's Priority tier requires the Interactions API and
-is not a valid `generateContent` option. NVIDIA NIM and local servers publish
+is not a valid `generateContent` option. NVIDIA Agentic and local servers publish
 no universal OpenRouter-style routing field, so UR does not invent one.
 API-key entry for
-OpenAI, Claude, Gemini, OpenRouter, NVIDIA NIM, and authenticated compatible
+OpenAI, Claude, Gemini, OpenRouter, NVIDIA Agentic/Special, and authenticated compatible
 endpoints is a single aligned masked row; the key is stored in the OS keychain
 flow and is never written to settings. On the model screen, `K` adds or
 replaces the selected HTTP provider's key and `E` edits its endpoint. Generic
@@ -313,7 +328,7 @@ UR uses each provider's non-generating count endpoint when one covers the full
 request: OpenAI Responses input tokens, Anthropic Messages token counting,
 Gemini `countTokens`, llama.cpp chat input tokens, and vLLM Messages token
 counting. NVIDIA's hosted NIM API has no documented token-count route;
-NVIDIA NIM, Ollama, OpenRouter, LM Studio, Unsloth, and subscription CLIs use a
+NVIDIA Agentic, Ollama, OpenRouter, LM Studio, Unsloth, and subscription CLIs use a
 provider-wire local estimate because those runtimes do not share a dependable
 preflight tokenizer for complete chat history plus tools. UR never launches a
 hidden completion for token counting. If a native count call is unavailable,
@@ -344,7 +359,7 @@ error, not a successful `Did 0 searches` result.
 When you select a UR-native provider and model, every agent request is routed
 through that provider's backend:
 
-- **API providers** make direct HTTP calls in each provider's native wire format: Anthropic uses `x-api-key` + `anthropic-version` against `/v1/messages`; OpenAI uses `Authorization: Bearer` against `/v1/chat/completions` by default or `/v1/responses` when explicitly selected; Gemini uses `x-goog-api-key` against `…:generateContent`; OpenRouter and NVIDIA NIM use their OpenAI-compatible chat endpoints.
+- **API providers** make direct calls in each provider's native wire format: Anthropic uses `x-api-key` + `anthropic-version` against `/v1/messages`; OpenAI uses `Authorization: Bearer` against `/v1/chat/completions` by default or `/v1/responses` when explicitly selected; Gemini uses `x-goog-api-key` against `…:generateContent`; OpenRouter and NVIDIA Agentic use documented chat endpoints; NVIDIA Special uses per-card HTTP, NVCF, or gRPC inference.
 - **Local/server providers** connect to the configured local or OpenAI-compatible endpoint (`/v1/chat/completions` for LM Studio, llama.cpp and vLLM; the native tags/chat API for Ollama)
 - **Subscription CLI providers** (Codex CLI, Claude Code, Gemini CLI,
   Antigravity) dispatch the turn through the vendor's official CLI using your
@@ -455,7 +470,8 @@ ur config set provider anthropic-api
 | --- | --- | --- |
 | API providers (openai-api, anthropic-api, gemini-api) | Live discovery from the provider's `/models` endpoint using your connected key (curated fallback until connected) | live |
 | OpenRouter | Live `/models` discovery with an endpoint-scoped five-minute cache; Ctrl+R forces a fresh request with no stale fallback | live/cache |
-| NVIDIA NIM | Hosted: live `/models` availability intersected with audited agent contracts, plus a generated official OpenAPI catalog for dedicated one-shot APIs. Configured NIM gateway: its own live `/models` catalog | live agents + official task contracts |
+| NVIDIA Agentic | Public Build: generated per-card agent contracts, never narrowed by account inventory. Configured NIM gateway: its own live `/models` catalog | live card contracts or gateway live |
+| NVIDIA Special | Generated from every current focused Free Endpoint card, including exact HTTP/NVCF/gRPC inference metadata | static generated contracts |
 | Local/server providers (ollama, lmstudio, llama.cpp, vllm, unsloth) | Dynamic discovery from the selected provider endpoint | live |
 | OpenAI-compatible | Dynamic discovery from configured endpoint | live |
 | Subscription CLIs (codex-cli, claude-code-cli, gemini-cli, antigravity-cli) | Curated list (the official CLIs expose no models API); first-class in `/model`, dispatched via the official CLI. External CLI behavior depends on the vendor CLI. Log in with `ur auth <provider>` | static |
@@ -479,10 +495,11 @@ provider's successful live catalog remains authoritative for that account.
 
 **API providers** require environment variable with API key:
 - `openai-api` — requires `OPENAI_API_KEY`
-- `anthropic-api` — requires `ANTHROPIC_API_KEY`
+- `anthropic-api` — requires `ANTHROPIC_API_KEY`; identity-linked keys also
+  select `ANTHROPIC_WORKSPACE_ID` (or saved `anthropic.workspace_id`)
 - `gemini-api` — requires `GEMINI_API_KEY`
 - `openrouter` — requires `OPENROUTER_API_KEY`
-- `nvidia-nim` — requires `NVIDIA_API_KEY` for build.nvidia.com; endpoint is configurable
+- `nvidia-nim` / `nvidia-special` — share `NVIDIA_API_KEY`; Agentic's enterprise endpoint is configurable, while Special routes each public model to its card-specific contract
 
 **Local/server providers** require local runtime or endpoint:
 - `ollama` — configurable local, LAN, or hosted Ollama server
@@ -616,6 +633,7 @@ Provider config and doctor commands accept canonical IDs and common aliases:
 | `gemini-api` | `gemini api`, `google gemini api` |
 | `openrouter` | `openrouter api` |
 | `nvidia-nim` | `nvidia`, `NVIDIA Build`, `nvidia api`, `nim` |
+| `nvidia-special` | `NVIDIA Special`, `nvidia task`, `nvidia one-shot` |
 | `openai-compatible` | `compatible`, `openai compatible` |
 | `ollama` | `ollama local` |
 | `lmstudio` | `LM Studio`, `lm-studio` |
@@ -654,6 +672,7 @@ API providers require explicit user selection and environment keys:
 OPENAI_API_KEY=...
 OPENAI_COMPATIBLE_API_KEY=...
 ANTHROPIC_API_KEY=...
+ANTHROPIC_WORKSPACE_ID=wrkspc_... # identity-linked Anthropic keys only
 GEMINI_API_KEY=...
 OPENROUTER_API_KEY=...
 NVIDIA_API_KEY=...
@@ -680,36 +699,35 @@ The compatible provider's key is optional and provider-scoped. Add or replace
 it with `ur connect openai-compatible`, `/connect openai-compatible`, or `K`
 in the `/model` model screen. Anonymous endpoints continue to work without it.
 
-### NVIDIA NIM / build.nvidia.com
+### NVIDIA Agentic and NVIDIA Special / build.nvidia.com
 
-NVIDIA NIM is a UR-native, OpenAI-compatible provider with live discovery:
+NVIDIA is exposed through two UR-native providers sharing one stored key:
 
 ```sh
 echo "$NVIDIA_API_KEY" | ur connect nvidia-nim
 ur config set provider nvidia-nim
 ur provider doctor nvidia-nim
+# Focused inference uses the same key:
+ur config set provider nvidia-special
 # Optional self-hosted/enterprise gateway:
 ur config set base_url nvidia-nim https://nim-gateway.example/v1
 ```
 
-The default is `https://integrate.api.nvidia.com/v1`. NVIDIA's authenticated
-`/v1/models` response proves current account availability but mixes agents,
-utilities, VLMs, and generation functions. UR intersects it with a reviewed
-positive agent registry before allowing a model to own the multi-turn tool
-loop. It does not intersect the result with NVCF's separate deployment-function
-inventory. A custom enterprise or self-hosted NIM remains independent and uses
-only that configured gateway's `/models` response.
-
-The Build web catalog also contains download-only NIMs. UR does not add those
-cards to the hosted picker: only IDs returned by the authenticated hosted
-`/v1/models` endpoint can appear. UR focuses
+NVIDIA Agentic defaults to `https://integrate.api.nvidia.com/v1`. For the
+public service, UR generates its agent catalog from every current Free Endpoint
+card that explicitly advertises agent/tool use. `/v1/models` is not an
+allowlist: account entitlement failures are reported and never remove a model.
+A custom enterprise or self-hosted NIM remains independent and uses only that
+configured gateway's live `/models` response. Download-only cards never enter
+either hosted provider. UR focuses
 `nvidia/nemotron-3.5-lightning-30b-a3b` first because NVIDIA documents it as
 its fastest 30B model for long-running agents. Left/Right can turn that exact
 model's thinking off/on through NVIDIA's documented
 `chat_template_kwargs.enable_thinking`; other NIM models receive only their
 own advertised reasoning contract.
 
-UR calls `/models` and `/chat/completions`. It counts NVIDIA requests with a
+Agentic uses each generated card's documented chat endpoint (or the configured
+gateway). It counts NVIDIA requests with a
 provider-wire local estimate instead of first calling the unsupported hosted
 `/messages/count_tokens` route, and never launches a hidden completion.
 Streaming, standard tool calls, and image input use the
@@ -719,32 +737,29 @@ documented Nemotron coding-agent models, UR includes NVIDIA's
 [NIM LLM API reference](https://docs.api.nvidia.com/nim/reference/llm-apis)
 and [NIM endpoint guide](https://docs.nvidia.com/nim/large-language-models/latest/tutorials.html).
 
-NVIDIA's dedicated APIs are a separate one-shot surface. `/model` labels them
-`ONE-SHOT`, shows each task's real purpose, and keeps the current agent model
-when one is selected. UR generates the executable catalog from NVIDIA's
-official LLM, retrieval, visual, multimodal, healthcare, route-optimization,
-and climate OpenAPI indexes. The current catalog has 92 tasks and routes each
-one to its documented endpoint on `integrate.api.nvidia.com`,
-`ai.api.nvidia.com`, `health.api.nvidia.com`, `optimize.api.nvidia.com`, or
-`climate.api.nvidia.com`.
-
-`NvidiaNimTask` accepts convenience prompt/image/query/passages fields, or its
-`describe` action exposes the exact request schema before an advanced
-`payload_json` call. `file_inputs` can bind local files into that payload by
-JSON pointer. UR automatically inlines small media or creates an NVIDIA Asset
-UUID/reference for larger and asset-based contracts, polls documented
-asynchronous responses, and saves binary or large JSON results under
-`.ur/artifacts/nvidia/`. It reuses `NVIDIA_API_KEY` and returns text/path
-metadata to the enclosing agent. Download-only cards, status routes,
-staging-only URLs, broken references, and operations without a documented
-public hosted POST endpoint never appear as usable choices.
-
-`ur provider doctor nvidia-nim` verifies the hosted agent catalog and selected
-agent model against live `/v1/models`, and reports the generated task-contract
-count. Dedicated API entitlement can be verified only with that task's valid
-payload. If NVIDIA rejects a listed model after selection, UR redacts internal
-function/account IDs and removes only that endpoint-scoped model until
-`Ctrl+R` explicitly retries discovery.
+NVIDIA Special is a separate focused-task provider. Selecting a row stores a
+one-shot preference and keeps the current chat provider/model. The generator
+crawls all 100 current Build cards and preserves every card labelled Free
+Endpoint: 36 total, split into 13 Agentic and 23 Special. Thirty-five publish a
+complete executable contract; 22 are Special. The remaining VoiceChat card is
+kept visible with `unpublished` transport because NVIDIA currently provides no
+public request/response protocol.
+
+Every executable row records the purpose, input/output hints, exact endpoint,
+HTTP or RPC method, card-specific function ID, request and response schemas,
+documentation, and current NVIDIA availability flag. HTTP cards use their own
+`integrate.api.nvidia.com`, `ai.api.nvidia.com`, or direct
+`{function-id}.invocation.api.nvcf.nvidia.com` route. Five Maxine/Riva cards use
+their exact public service/method on `grpc.nvcf.nvidia.com:443`.
+
+`NvidiaSpecial` exposes `describe` for the full contract and `run` for
+convenience text/image/audio/video inputs or exact `payload_json` with
+JSON-pointer file bindings. UR inlines small media, uses NVIDIA Assets for
+larger inputs, polls documented asynchronous requests, decodes single or
+multiple returned artifacts, and saves output under `.ur/artifacts/nvidia/`
+unless the caller supplies a path. Both providers reuse `NVIDIA_API_KEY`.
+NVIDIA availability and entitlement errors are redacted and reported, but no
+model is ever removed from either generated catalog.
 
 Local/server providers use their normal endpoints:
 
@@ -802,10 +817,10 @@ Required variables:
 | --- | --- | --- |
 | OpenAI-compatible | `OPENAI_COMPATIBLE_BASE_URL`, `OPENAI_COMPATIBLE_MODEL` | `OPENAI_COMPATIBLE_API_KEY` |
 | Unsloth | `UNSLOTH_API_KEY`, `UNSLOTH_MODEL` | `UNSLOTH_BASE_URL` (defaults to `http://localhost:8888/v1`) |
-| NVIDIA NIM | `NVIDIA_API_KEY`, `NVIDIA_MODEL` | `NVIDIA_BASE_URL` (defaults to `https://integrate.api.nvidia.com/v1`) |
+| NVIDIA Agentic | `NVIDIA_API_KEY`, `NVIDIA_MODEL` | `NVIDIA_BASE_URL` (defaults to `https://integrate.api.nvidia.com/v1`) |
 | OpenAI | `OPENAI_API_KEY`, `OPENAI_MODEL` | `OPENAI_BASE_URL` |
 | OpenRouter | `OPENROUTER_API_KEY`, `OPENROUTER_MODEL` | `OPENROUTER_BASE_URL` |
-| Anthropic | `ANTHROPIC_API_KEY`, `ANTHROPIC_MODEL` | `ANTHROPIC_BASE_URL` |
+| Anthropic | `ANTHROPIC_API_KEY`, `ANTHROPIC_MODEL` | `ANTHROPIC_BASE_URL`; `ANTHROPIC_WORKSPACE_ID` for identity-linked keys |
 | Gemini | `GEMINI_API_KEY`, `GEMINI_MODEL` | `GEMINI_BASE_URL` |
 | Ollama | `OLLAMA_MODEL` | `OLLAMA_BASE_URL` or `OLLAMA_HOST`; `OLLAMA_API_KEY` when required |
 | LM Studio | `LMSTUDIO_BASE_URL`, `LMSTUDIO_MODEL` | `LMSTUDIO_API_KEY` |

package/documentation/app.js

@@ -68,8 +68,8 @@ const featureGroups = [
   {
     title: 'Providers and auth',
     tags: ['subscription', 'API', 'local', 'effort', 'status bar'],
-    text: 'UR-native API/local/OpenAI-compatible runtimes, provider-scoped endpoints, audited NVIDIA agent discovery plus a generated official OpenAPI catalog for 92 exact one-shot AI/retrieval/health/optimization/climate contracts, provider-only Unsloth inference, optional compatible-gateway keys, capability-driven reasoning effort, responsive OpenRouter routing, first-class subscription CLI providers dispatched through the official vendor CLIs, provider doctor checks, secure API-key connect, non-secret config, fallback hints, and provider-aware status-bar output.',
-    commands: ['ur provider list', 'ur provider status', 'ur provider doctor nvidia-nim', 'ur connect status', 'ur config set provider nvidia-nim', 'ur config set provider openai-api', 'ur config set provider ollama', 'ur config set base_url llama.cpp http://localhost:9931/v1', '/model', '/effort ultra', '/thinking on'],
+    text: 'UR-native API/local/OpenAI-compatible runtimes, provider-scoped endpoints, NVIDIA Agentic plus NVIDIA Special with all 36 current Build Free Endpoint cards and exact per-card HTTP/NVCF/gRPC inference contracts, provider-only Unsloth inference, optional compatible-gateway keys, capability-driven reasoning effort, responsive OpenRouter routing, first-class subscription CLI providers dispatched through official vendor CLIs, provider doctor checks, secure API-key connect, non-secret config, fallback hints, and provider-aware status-bar output.',
+    commands: ['ur provider list', 'ur provider status', 'ur provider doctor nvidia-nim', 'ur connect status', 'ur config set provider nvidia-nim', 'ur config set provider nvidia-special', 'ur config set provider openai-api', 'ur config set provider ollama', 'ur config set base_url llama.cpp http://localhost:9931/v1', '/model', '/effort ultra', '/thinking on'],
   },
   {
     title: 'Security and operations',

package/documentation/index.html

@@ -45,7 +45,7 @@
     <main id="content" class="content">
       <header class="topbar">
         <div>
-          <p class="eyebrow">Version 1.84.7</p>
+          <p class="eyebrow">Version 1.85.1</p>
           <h1>UR-Nexus Documentation</h1>
           <p class="lead">A practical, tutorial-style reference for installing, configuring, automating, extending, and operating UR-Nexus.</p>
         </div>
@@ -166,6 +166,7 @@ ur provider doctor agy</code></pre>
             <pre><code>ur config set provider openai-compatible
 ur config set provider openai-api
 ur config set provider nvidia-nim
+ur config set provider nvidia-special
 ur config set base_url ollama http://localhost:11434
 ur config set base_url llama.cpp http://localhost:9931/v1
 ur config set provider unsloth
@@ -174,7 +175,7 @@ ur config set provider.fallback ollama
 ur config set openai_transport responses
 ur config set responses.store false
 ur config set anthropic.speed fast</code></pre>
-            <p>API providers require explicit selection and read keys from a key stored via <code>ur connect</code> (OS keychain) or from environment variables. Each configurable provider keeps its own <code>base_url</code>, so switching among Ollama, LM Studio, llama.cpp, vLLM, Unsloth, NVIDIA NIM, and API gateways restores the matching address. OpenAI Responses is opt-in and privacy-first; Chat Completions remains the default. Direct Anthropic preserves prompt-cache breakpoints, streams tool inputs with its per-tool control, and enables the documented premium fast tier only when explicitly configured for a supported model. Unsloth is an authenticated inference provider only.</p>
+            <p>API providers require explicit selection and read keys from a key stored via <code>ur connect</code> (OS keychain) or from environment variables. Each configurable provider keeps its own <code>base_url</code>, so switching among Ollama, LM Studio, llama.cpp, vLLM, Unsloth, NVIDIA Agentic, and API gateways restores the matching address. OpenAI Responses is opt-in and privacy-first; Chat Completions remains the default. Direct Anthropic preserves prompt-cache breakpoints, streams tool inputs with its per-tool control, and enables the documented premium fast tier only when explicitly configured for a supported model. Unsloth is an authenticated inference provider only.</p>
           </article>
           <article>
             <h3>Capability-driven reasoning effort</h3>
@@ -191,12 +192,13 @@ ur --model kimi-k3:cloud --effort high
             <p>The endpoint-scoped model catalog is reused for five minutes, while forced refresh never substitutes stale data. Tool turns preserve OpenRouter Auto Exacto for live throughput and tool-call reliability; non-tool turns prefer end-to-end throughput. Stable session affinity keeps prompt caches warm, while configurable routing preferences and the <code>:nitro</code>, <code>:floor</code>, and <code>:exacto</code> variants remain authoritative.</p>
           </article>
           <article>
-            <h3>NVIDIA NIM and compatible API keys</h3>
+            <h3>NVIDIA Agentic, NVIDIA Special, and compatible API keys</h3>
             <pre><code>ur connect nvidia-nim
 ur config set provider nvidia-nim
+ur config set provider nvidia-special
 ur config set base_url nvidia-nim https://integrate.api.nvidia.com/v1
 /model  # K API key · E endpoint</code></pre>
-            <p>NVIDIA NIM separates ongoing agents from specialized one-shot models. Hosted agent discovery intersects live <code>/v1/models</code> inventory with audited tool-loop contracts. The separately labelled one-shot catalog is generated from NVIDIA's official public OpenAPI indexes and currently covers 92 executable AI, retrieval, health, optimization, and climate contracts. Each entry shows its purpose and routes to its exact documented endpoint with the same stored NVIDIA key; schema validation, NVIDIA Asset upload, asynchronous polling, and binary/large-JSON artifact saving are built in. Download-only, staging-only, broken, and non-executable operations stay hidden, and choosing a task never changes the ongoing agent. Nemotron 3.5 Lightning retains its documented on/off thinking field; unknown models inherit no fabricated effort. Generic OpenAI-compatible endpoints can store an optional dedicated key, while anonymous endpoints remain valid.</p>
+            <p>NVIDIA is split into NVIDIA Agentic for the ongoing tool loop and NVIDIA Special for focused inference; both share one stored <code>NVIDIA_API_KEY</code>. UR audits all 100 current Build cards and preserves every Free Endpoint card: 13 Agentic and 23 Special. Thirty-five publish executable contracts; the one unpublished protocol stays visible and labelled. Each row shows purpose, input, and output, and uses that card's exact HTTP, direct NVCF, or gRPC endpoint, method, function ID, and request/response schema. NVIDIA Assets, asynchronous polling, five native Maxine/Riva gRPC services, and media/binary/JSON artifact saving are built in. Account errors never remove models, and choosing a Special task never changes the ongoing agent. Nemotron 3.5 Lightning retains its documented on/off thinking field; unknown models inherit no fabricated effort. Generic OpenAI-compatible endpoints can store an optional dedicated key, while anonymous endpoints remain valid.</p>
           </article>
           <article>
             <h3>Portable shell deadlines</h3>
@@ -207,7 +209,7 @@ ur config set base_url nvidia-nim https://integrate.api.nvidia.com/v1
           <article>
             <h3>Tool images across providers</h3>
             <pre><code>Computer screenshot → tool result → vision model</code></pre>
-            <p>Every UR-native provider preserves image-bearing tool output in its valid wire shape. OpenRouter, NVIDIA NIM, and other Chat-Completions backends keep the tool result textual and send the image in the next multimodal user turn; Gemini, OpenAI Responses, Anthropic, and Ollama use their native rich forms. Vision still depends on the selected model.</p>
+            <p>Every UR-native provider preserves image-bearing tool output in its valid wire shape. OpenRouter, NVIDIA Agentic, and other Chat-Completions backends keep the tool result textual and send the image in the next multimodal user turn; Gemini, OpenAI Responses, Anthropic, and Ollama use their native rich forms. Vision still depends on the selected model.</p>
           </article>
           <article>
             <h3>Status bar and updates</h3>

package/extensions/jetbrains-ur/build.gradle.kts

@@ -7,7 +7,7 @@ plugins {
 }
 
 group = "dev.urnexus"
-version = "1.84.7"
+version = "1.85.1"
 
 repositories {
     mavenCentral()

package/extensions/vscode-ur-inline-diffs/package.json

@@ -2,7 +2,7 @@
   "name": "ur-inline-diffs",
   "displayName": "UR Inline Diffs",
   "description": "Review, apply, and reject UR inline diff bundles from .ur/ide/diffs inside VS Code.",
-  "version": "1.84.7",
+  "version": "1.85.1",
   "publisher": "ur-nexus",
   "engines": {
     "vscode": "^1.92.0"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ur-agent",
-  "version": "1.84.7",
+  "version": "1.85.1",
   "description": "UR-Nexus — autonomous engineering workflow engine (plan, execute, test, verify, document, benchmark, reproduce)",
   "type": "module",
   "packageManager": "bun@1.3.14",
@@ -86,6 +86,8 @@
     "@ag-ui/core": "0.0.57",
     "@ag-ui/encoder": "0.0.57",
     "@anthropic-ai/sandbox-runtime": "0.0.71",
+    "@grpc/grpc-js": "^1.14.0",
+    "@grpc/proto-loader": "^0.8.0",
     "diff2html": "^3.4.56",
     "playwright-core": "^1.61.1",
     "sharp": "^0.35.3"