durar-ai 2026.4.4

package/docs/providers/synthetic.md

@@ -0,0 +1,101 @@
+---
+summary: "Use Synthetic's Anthropic-compatible API in Durar"
+read_when:
+  - You want to use Synthetic as a model provider
+  - You need a Synthetic API key or base URL setup
+title: "Synthetic"
+---
+
+# Synthetic
+
+Synthetic exposes Anthropic-compatible endpoints. Durar registers it as the
+`synthetic` provider and uses the Anthropic Messages API.
+
+## Quick setup
+
+1. Set `SYNTHETIC_API_KEY` (or run the wizard below).
+2. Run onboarding:
+
+```bash
+Durar onboard --auth-choice synthetic-api-key
+```
+
+The default model is set to:
+
+```
+synthetic/hf:MiniMaxAI/MiniMax-M2.5
+```
+
+## Config example
+
+```json5
+{
+  env: { SYNTHETIC_API_KEY: "sk-..." },
+  agents: {
+    defaults: {
+      model: { primary: "synthetic/hf:MiniMaxAI/MiniMax-M2.5" },
+      models: { "synthetic/hf:MiniMaxAI/MiniMax-M2.5": { alias: "MiniMax M2.5" } },
+    },
+  },
+  models: {
+    mode: "merge",
+    providers: {
+      synthetic: {
+        baseUrl: "https://api.synthetic.new/anthropic",
+        apiKey: "${SYNTHETIC_API_KEY}",
+        api: "anthropic-messages",
+        models: [
+          {
+            id: "hf:MiniMaxAI/MiniMax-M2.5",
+            name: "MiniMax M2.5",
+            reasoning: false,
+            input: ["text"],
+            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+            contextWindow: 192000,
+            maxTokens: 65536,
+          },
+        ],
+      },
+    },
+  },
+}
+```
+
+Note: Durar's Anthropic client appends `/v1` to the base URL, so use
+`https://api.synthetic.new/anthropic` (not `/anthropic/v1`). If Synthetic changes
+its base URL, override `models.providers.synthetic.baseUrl`.
+
+## Model catalog
+
+All models below use cost `0` (input/output/cache).
+
+| Model ID                                               | Context window | Max tokens | Reasoning | Input        |
+| ------------------------------------------------------ | -------------- | ---------- | --------- | ------------ |
+| `hf:MiniMaxAI/MiniMax-M2.5`                            | 192000         | 65536      | false     | text         |
+| `hf:moonshotai/Kimi-K2-Thinking`                       | 256000         | 8192       | true      | text         |
+| `hf:zai-org/GLM-4.7`                                   | 198000         | 128000     | false     | text         |
+| `hf:deepseek-ai/DeepSeek-R1-0528`                      | 128000         | 8192       | false     | text         |
+| `hf:deepseek-ai/DeepSeek-V3-0324`                      | 128000         | 8192       | false     | text         |
+| `hf:deepseek-ai/DeepSeek-V3.1`                         | 128000         | 8192       | false     | text         |
+| `hf:deepseek-ai/DeepSeek-V3.1-Terminus`                | 128000         | 8192       | false     | text         |
+| `hf:deepseek-ai/DeepSeek-V3.2`                         | 159000         | 8192       | false     | text         |
+| `hf:meta-llama/Llama-3.3-70B-Instruct`                 | 128000         | 8192       | false     | text         |
+| `hf:meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8` | 524000         | 8192       | false     | text         |
+| `hf:moonshotai/Kimi-K2-Instruct-0905`                  | 256000         | 8192       | false     | text         |
+| `hf:moonshotai/Kimi-K2.5`                              | 256000         | 8192       | true      | text + image |
+| `hf:openai/gpt-oss-120b`                               | 128000         | 8192       | false     | text         |
+| `hf:Qwen/Qwen3-235B-A22B-Instruct-2507`                | 256000         | 8192       | false     | text         |
+| `hf:Qwen/Qwen3-Coder-480B-A35B-Instruct`               | 256000         | 8192       | false     | text         |
+| `hf:Qwen/Qwen3-VL-235B-A22B-Instruct`                  | 250000         | 8192       | false     | text + image |
+| `hf:zai-org/GLM-4.5`                                   | 128000         | 128000     | false     | text         |
+| `hf:zai-org/GLM-4.6`                                   | 198000         | 128000     | false     | text         |
+| `hf:zai-org/GLM-5`                                     | 256000         | 128000     | true      | text + image |
+| `hf:deepseek-ai/DeepSeek-V3`                           | 128000         | 8192       | false     | text         |
+| `hf:Qwen/Qwen3-235B-A22B-Thinking-2507`                | 256000         | 8192       | true      | text         |
+
+## Notes
+
+- Model refs use `synthetic/<modelId>`.
+- If you enable a model allowlist (`agents.defaults.models`), add every model you
+  plan to use.
+- See [Model providers](/concepts/model-providers) for provider rules.

package/docs/providers/together.md

@@ -0,0 +1,70 @@
+---
+title: "Together AI"
+summary: "Together AI setup (auth + model selection)"
+read_when:
+  - You want to use Together AI with Durar
+  - You need the API key env var or CLI auth choice
+---
+
+# Together AI
+
+The [Together AI](https://together.ai) provides access to leading open-source models including Llama, DeepSeek, Kimi, and more through a unified API.
+
+- Provider: `together`
+- Auth: `TOGETHER_API_KEY`
+- API: OpenAI-compatible
+- Base URL: `https://api.together.xyz/v1`
+
+## Quick start
+
+1. Set the API key (recommended: store it for the Gateway):
+
+```bash
+Durar onboard --auth-choice together-api-key
+```
+
+2. Set a default model:
+
+```json5
+{
+  agents: {
+    defaults: {
+      model: { primary: "together/moonshotai/Kimi-K2.5" },
+    },
+  },
+}
+```
+
+## Non-interactive example
+
+```bash
+Durar onboard --non-interactive \
+  --mode local \
+  --auth-choice together-api-key \
+  --together-api-key "$TOGETHER_API_KEY"
+```
+
+This will set `together/moonshotai/Kimi-K2.5` as the default model.
+
+## Environment note
+
+If the Gateway runs as a daemon (launchd/systemd), make sure `TOGETHER_API_KEY`
+is available to that process (for example, in `~/.Durar/.env` or via
+`env.shellEnv`).
+
+## Built-in catalog
+
+Durar currently ships this bundled Together catalog:
+
+| Model ref                                                    | Name                                   | Input       | Context    | Notes                            |
+| ------------------------------------------------------------ | -------------------------------------- | ----------- | ---------- | -------------------------------- |
+| `together/moonshotai/Kimi-K2.5`                              | Kimi K2.5                              | text, image | 262,144    | Default model; reasoning enabled |
+| `together/zai-org/GLM-4.7`                                   | GLM 4.7 Fp8                            | text        | 202,752    | General-purpose text model       |
+| `together/meta-llama/Llama-3.3-70B-Instruct-Turbo`           | Llama 3.3 70B Instruct Turbo           | text        | 131,072    | Fast instruction model           |
+| `together/meta-llama/Llama-4-Scout-17B-16E-Instruct`         | Llama 4 Scout 17B 16E Instruct         | text, image | 10,000,000 | Multimodal                       |
+| `together/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8` | Llama 4 Maverick 17B 128E Instruct FP8 | text, image | 20,000,000 | Multimodal                       |
+| `together/deepseek-ai/DeepSeek-V3.1`                         | DeepSeek V3.1                          | text        | 131,072    | General text model               |
+| `together/deepseek-ai/DeepSeek-R1`                           | DeepSeek R1                            | text        | 131,072    | Reasoning model                  |
+| `together/moonshotai/Kimi-K2-Instruct-0905`                  | Kimi K2-Instruct 0905                  | text        | 262,144    | Secondary Kimi text model        |
+
+The onboarding preset sets `together/moonshotai/Kimi-K2.5` as the default model.

package/docs/providers/venice.md

@@ -0,0 +1,282 @@
+---
+summary: "Use Venice AI privacy-focused models in Durar"
+read_when:
+  - You want privacy-focused inference in Durar
+  - You want Venice AI setup guidance
+title: "Venice AI"
+---
+
+# Venice AI (Venice highlight)
+
+**Venice** is our highlight Venice setup for privacy-first inference with optional anonymized access to proprietary models.
+
+Venice AI provides privacy-focused AI inference with support for uncensored models and access to major proprietary models through their anonymized proxy. All inference is private by default—no training on your data, no logging.
+
+## Why Venice in Durar
+
+- **Private inference** for open-source models (no logging).
+- **Uncensored models** when you need them.
+- **Anonymized access** to proprietary models (Opus/GPT/Gemini) when quality matters.
+- OpenAI-compatible `/v1` endpoints.
+
+## Privacy Modes
+
+Venice offers two privacy levels — understanding this is key to choosing your model:
+
+| Mode           | Description                                                                                                                       | Models                                                        |
+| -------------- | --------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------- |
+| **Private**    | Fully private. Prompts/responses are **never stored or logged**. Ephemeral.                                                       | Llama, Qwen, DeepSeek, Kimi, MiniMax, Venice Uncensored, etc. |
+| **Anonymized** | Proxied through Venice with metadata stripped. The underlying provider (OpenAI, Anthropic, Google, xAI) sees anonymized requests. | Claude, GPT, Gemini, Grok                                     |
+
+## Features
+
+- **Privacy-focused**: Choose between "private" (fully private) and "anonymized" (proxied) modes
+- **Uncensored models**: Access to models without content restrictions
+- **Major model access**: Use Claude, GPT, Gemini, and Grok via Venice's anonymized proxy
+- **OpenAI-compatible API**: Standard `/v1` endpoints for easy integration
+- **Streaming**: ✅ Supported on all models
+- **Function calling**: ✅ Supported on select models (check model capabilities)
+- **Vision**: ✅ Supported on models with vision capability
+- **No hard rate limits**: Fair-use throttling may apply for extreme usage
+
+## Setup
+
+### 1. Get API Key
+
+1. Sign up at [venice.ai](https://venice.ai)
+2. Go to **Settings → API Keys → Create new key**
+3. Copy your API key (format: `vapi_xxxxxxxxxxxx`)
+
+### 2. Configure Durar
+
+**Option A: Environment Variable**
+
+```bash
+export VENICE_API_KEY="vapi_xxxxxxxxxxxx"
+```
+
+**Option B: Interactive Setup (Recommended)**
+
+```bash
+Durar onboard --auth-choice venice-api-key
+```
+
+This will:
+
+1. Prompt for your API key (or use existing `VENICE_API_KEY`)
+2. Show all available Venice models
+3. Let you pick your default model
+4. Configure the provider automatically
+
+**Option C: Non-interactive**
+
+```bash
+Durar onboard --non-interactive \
+  --auth-choice venice-api-key \
+  --venice-api-key "vapi_xxxxxxxxxxxx"
+```
+
+### 3. Verify Setup
+
+```bash
+Durar agent --model venice/kimi-k2-5 --message "Hello, are you working?"
+```
+
+## Model Selection
+
+After setup, Durar shows all available Venice models. Pick based on your needs:
+
+- **Default model**: `venice/kimi-k2-5` for strong private reasoning plus vision.
+- **High-capability option**: `venice/claude-opus-4-6` for the strongest anonymized Venice path.
+- **Privacy**: Choose "private" models for fully private inference.
+- **Capability**: Choose "anonymized" models to access Claude, GPT, Gemini via Venice's proxy.
+
+Change your default model anytime:
+
+```bash
+Durar models set venice/kimi-k2-5
+Durar models set venice/claude-opus-4-6
+```
+
+List all available models:
+
+```bash
+Durar models list | grep venice
+```
+
+## Configure via `Durar configure`
+
+1. Run `Durar configure`
+2. Select **Model/auth**
+3. Choose **Venice AI**
+
+## Which Model Should I Use?
+
+| Use Case                   | Recommended Model                | Why                                          |
+| -------------------------- | -------------------------------- | -------------------------------------------- |
+| **General chat (default)** | `kimi-k2-5`                      | Strong private reasoning plus vision         |
+| **Best overall quality**   | `claude-opus-4-6`                | Strongest anonymized Venice option           |
+| **Privacy + coding**       | `qwen3-coder-480b-a35b-instruct` | Private coding model with large context      |
+| **Private vision**         | `kimi-k2-5`                      | Vision support without leaving private mode  |
+| **Fast + cheap**           | `qwen3-4b`                       | Lightweight reasoning model                  |
+| **Complex private tasks**  | `deepseek-v3.2`                  | Strong reasoning, but no Venice tool support |
+| **Uncensored**             | `venice-uncensored`              | No content restrictions                      |
+
+## Available Models (41 Total)
+
+### Private Models (26) - Fully Private, No Logging
+
+| Model ID                               | Name                                | Context | Features                   |
+| -------------------------------------- | ----------------------------------- | ------- | -------------------------- |
+| `kimi-k2-5`                            | Kimi K2.5                           | 256k    | Default, reasoning, vision |
+| `kimi-k2-thinking`                     | Kimi K2 Thinking                    | 256k    | Reasoning                  |
+| `llama-3.3-70b`                        | Llama 3.3 70B                       | 128k    | General                    |
+| `llama-3.2-3b`                         | Llama 3.2 3B                        | 128k    | General                    |
+| `hermes-3-llama-3.1-405b`              | Hermes 3 Llama 3.1 405B             | 128k    | General, tools disabled    |
+| `qwen3-235b-a22b-thinking-2507`        | Qwen3 235B Thinking                 | 128k    | Reasoning                  |
+| `qwen3-235b-a22b-instruct-2507`        | Qwen3 235B Instruct                 | 128k    | General                    |
+| `qwen3-coder-480b-a35b-instruct`       | Qwen3 Coder 480B                    | 256k    | Coding                     |
+| `qwen3-coder-480b-a35b-instruct-turbo` | Qwen3 Coder 480B Turbo              | 256k    | Coding                     |
+| `qwen3-5-35b-a3b`                      | Qwen3.5 35B A3B                     | 256k    | Reasoning, vision          |
+| `qwen3-next-80b`                       | Qwen3 Next 80B                      | 256k    | General                    |
+| `qwen3-vl-235b-a22b`                   | Qwen3 VL 235B (Vision)              | 256k    | Vision                     |
+| `qwen3-4b`                             | Venice Small (Qwen3 4B)             | 32k     | Fast, reasoning            |
+| `deepseek-v3.2`                        | DeepSeek V3.2                       | 160k    | Reasoning, tools disabled  |
+| `venice-uncensored`                    | Venice Uncensored (Dolphin-Mistral) | 32k     | Uncensored, tools disabled |
+| `mistral-31-24b`                       | Venice Medium (Mistral)             | 128k    | Vision                     |
+| `google-gemma-3-27b-it`                | Google Gemma 3 27B Instruct         | 198k    | Vision                     |
+| `openai-gpt-oss-120b`                  | OpenAI GPT OSS 120B                 | 128k    | General                    |
+| `nvidia-nemotron-3-nano-30b-a3b`       | NVIDIA Nemotron 3 Nano 30B          | 128k    | General                    |
+| `olafangensan-glm-4.7-flash-heretic`   | GLM 4.7 Flash Heretic               | 128k    | Reasoning                  |
+| `zai-org-glm-4.6`                      | GLM 4.6                             | 198k    | General                    |
+| `zai-org-glm-4.7`                      | GLM 4.7                             | 198k    | Reasoning                  |
+| `zai-org-glm-4.7-flash`                | GLM 4.7 Flash                       | 128k    | Reasoning                  |
+| `zai-org-glm-5`                        | GLM 5                               | 198k    | Reasoning                  |
+| `minimax-m21`                          | MiniMax M2.1                        | 198k    | Reasoning                  |
+| `minimax-m25`                          | MiniMax M2.5                        | 198k    | Reasoning                  |
+
+### Anonymized Models (15) - Via Venice Proxy
+
+| Model ID                        | Name                           | Context | Features                  |
+| ------------------------------- | ------------------------------ | ------- | ------------------------- |
+| `claude-opus-4-6`               | Claude Opus 4.6 (via Venice)   | 1M      | Reasoning, vision         |
+| `claude-opus-4-5`               | Claude Opus 4.5 (via Venice)   | 198k    | Reasoning, vision         |
+| `claude-sonnet-4-6`             | Claude Sonnet 4.6 (via Venice) | 1M      | Reasoning, vision         |
+| `claude-sonnet-4-5`             | Claude Sonnet 4.5 (via Venice) | 198k    | Reasoning, vision         |
+| `openai-gpt-54`                 | GPT-5.4 (via Venice)           | 1M      | Reasoning, vision         |
+| `openai-gpt-53-codex`           | GPT-5.3 Codex (via Venice)     | 400k    | Reasoning, vision, coding |
+| `openai-gpt-52`                 | GPT-5.2 (via Venice)           | 256k    | Reasoning                 |
+| `openai-gpt-52-codex`           | GPT-5.2 Codex (via Venice)     | 256k    | Reasoning, vision, coding |
+| `openai-gpt-4o-2024-11-20`      | GPT-4o (via Venice)            | 128k    | Vision                    |
+| `openai-gpt-4o-mini-2024-07-18` | GPT-4o Mini (via Venice)       | 128k    | Vision                    |
+| `gemini-3-1-pro-preview`        | Gemini 3.1 Pro (via Venice)    | 1M      | Reasoning, vision         |
+| `gemini-3-pro-preview`          | Gemini 3 Pro (via Venice)      | 198k    | Reasoning, vision         |
+| `gemini-3-flash-preview`        | Gemini 3 Flash (via Venice)    | 256k    | Reasoning, vision         |
+| `grok-41-fast`                  | Grok 4.1 Fast (via Venice)     | 1M      | Reasoning, vision         |
+| `grok-code-fast-1`              | Grok Code Fast 1 (via Venice)  | 256k    | Reasoning, coding         |
+
+## Model Discovery
+
+Durar automatically discovers models from the Venice API when `VENICE_API_KEY` is set. If the API is unreachable, it falls back to a static catalog.
+
+The `/models` endpoint is public (no auth needed for listing), but inference requires a valid API key.
+
+## Streaming & Tool Support
+
+| Feature              | Support                                                 |
+| -------------------- | ------------------------------------------------------- |
+| **Streaming**        | ✅ All models                                           |
+| **Function calling** | ✅ Most models (check `supportsFunctionCalling` in API) |
+| **Vision/Images**    | ✅ Models marked with "Vision" feature                  |
+| **JSON mode**        | ✅ Supported via `response_format`                      |
+
+## Pricing
+
+Venice uses a credit-based system. Check [venice.ai/pricing](https://venice.ai/pricing) for current rates:
+
+- **Private models**: Generally lower cost
+- **Anonymized models**: Similar to direct API pricing + small Venice fee
+
+## Comparison: Venice vs Direct API
+
+| Aspect       | Venice (Anonymized)           | Direct API          |
+| ------------ | ----------------------------- | ------------------- |
+| **Privacy**  | Metadata stripped, anonymized | Your account linked |
+| **Latency**  | +10-50ms (proxy)              | Direct              |
+| **Features** | Most features supported       | Full features       |
+| **Billing**  | Venice credits                | Provider billing    |
+
+## Usage Examples
+
+```bash
+# Use the default private model
+Durar agent --model venice/kimi-k2-5 --message "Quick health check"
+
+# Use Claude Opus via Venice (anonymized)
+Durar agent --model venice/claude-opus-4-6 --message "Summarize this task"
+
+# Use uncensored model
+Durar agent --model venice/venice-uncensored --message "Draft options"
+
+# Use vision model with image
+Durar agent --model venice/qwen3-vl-235b-a22b --message "Review attached image"
+
+# Use coding model
+Durar agent --model venice/qwen3-coder-480b-a35b-instruct --message "Refactor this function"
+```
+
+## Troubleshooting
+
+### API key not recognized
+
+```bash
+echo $VENICE_API_KEY
+Durar models list | grep venice
+```
+
+Ensure the key starts with `vapi_`.
+
+### Model not available
+
+The Venice model catalog updates dynamically. Run `Durar models list` to see currently available models. Some models may be temporarily offline.
+
+### Connection issues
+
+Venice API is at `https://api.venice.ai/api/v1`. Ensure your network allows HTTPS connections.
+
+## Config file example
+
+```json5
+{
+  env: { VENICE_API_KEY: "vapi_..." },
+  agents: { defaults: { model: { primary: "venice/kimi-k2-5" } } },
+  models: {
+    mode: "merge",
+    providers: {
+      venice: {
+        baseUrl: "https://api.venice.ai/api/v1",
+        apiKey: "${VENICE_API_KEY}",
+        api: "openai-completions",
+        models: [
+          {
+            id: "kimi-k2-5",
+            name: "Kimi K2.5",
+            reasoning: true,
+            input: ["text", "image"],
+            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+            contextWindow: 256000,
+            maxTokens: 65536,
+          },
+        ],
+      },
+    },
+  },
+}
+```
+
+## Links
+
+- [Venice AI](https://venice.ai)
+- [API Documentation](https://docs.venice.ai)
+- [Pricing](https://venice.ai/pricing)
+- [Status](https://status.venice.ai)

package/docs/providers/vercel-ai-gateway.md

@@ -0,0 +1,60 @@
+---
+title: "Vercel AI Gateway"
+summary: "Vercel AI Gateway setup (auth + model selection)"
+read_when:
+  - You want to use Vercel AI Gateway with Durar
+  - You need the API key env var or CLI auth choice
+---
+
+# Vercel AI Gateway
+
+The [Vercel AI Gateway](https://vercel.com/ai-gateway) provides a unified API to access hundreds of models through a single endpoint.
+
+- Provider: `vercel-ai-gateway`
+- Auth: `AI_GATEWAY_API_KEY`
+- API: Anthropic Messages compatible
+- Durar auto-discovers the Gateway `/v1/models` catalog, so `/models vercel-ai-gateway`
+  includes current model refs such as `vercel-ai-gateway/openai/gpt-5.4`.
+
+## Quick start
+
+1. Set the API key (recommended: store it for the Gateway):
+
+```bash
+Durar onboard --auth-choice ai-gateway-api-key
+```
+
+2. Set a default model:
+
+```json5
+{
+  agents: {
+    defaults: {
+      model: { primary: "vercel-ai-gateway/anthropic/claude-opus-4.6" },
+    },
+  },
+}
+```
+
+## Non-interactive example
+
+```bash
+Durar onboard --non-interactive \
+  --mode local \
+  --auth-choice ai-gateway-api-key \
+  --ai-gateway-api-key "$AI_GATEWAY_API_KEY"
+```
+
+## Environment note
+
+If the Gateway runs as a daemon (launchd/systemd), make sure `AI_GATEWAY_API_KEY`
+is available to that process (for example, in `~/.Durar/.env` or via
+`env.shellEnv`).
+
+## Model ID shorthand
+
+Durar accepts Vercel Claude shorthand model refs and normalizes them at
+runtime:
+
+- `vercel-ai-gateway/claude-opus-4.6` -> `vercel-ai-gateway/anthropic/claude-opus-4.6`
+- `vercel-ai-gateway/opus-4.6` -> `vercel-ai-gateway/anthropic/claude-opus-4-6`

package/docs/providers/vllm.md

@@ -0,0 +1,103 @@
+---
+summary: "Run Durar with vLLM (OpenAI-compatible local server)"
+read_when:
+  - You want to run Durar against a local vLLM server
+  - You want OpenAI-compatible /v1 endpoints with your own models
+title: "vLLM"
+---
+
+# vLLM
+
+vLLM can serve open-source (and some custom) models via an **OpenAI-compatible** HTTP API. Durar can connect to vLLM using the `openai-completions` API.
+
+Durar can also **auto-discover** available models from vLLM when you opt in with `VLLM_API_KEY` (any value works if your server doesn’t enforce auth) and you do not define an explicit `models.providers.vllm` entry.
+
+## Quick start
+
+1. Start vLLM with an OpenAI-compatible server.
+
+Your base URL should expose `/v1` endpoints (e.g. `/v1/models`, `/v1/chat/completions`). vLLM commonly runs on:
+
+- `http://127.0.0.1:8000/v1`
+
+2. Opt in (any value works if no auth is configured):
+
+```bash
+export VLLM_API_KEY="vllm-local"
+```
+
+3. Select a model (replace with one of your vLLM model IDs):
+
+```json5
+{
+  agents: {
+    defaults: {
+      model: { primary: "vllm/your-model-id" },
+    },
+  },
+}
+```
+
+## Model discovery (implicit provider)
+
+When `VLLM_API_KEY` is set (or an auth profile exists) and you **do not** define `models.providers.vllm`, Durar will query:
+
+- `GET http://127.0.0.1:8000/v1/models`
+
+…and convert the returned IDs into model entries.
+
+If you set `models.providers.vllm` explicitly, auto-discovery is skipped and you must define models manually.
+
+## Explicit configuration (manual models)
+
+Use explicit config when:
+
+- vLLM runs on a different host/port.
+- You want to pin `contextWindow`/`maxTokens` values.
+- Your server requires a real API key (or you want to control headers).
+
+```json5
+{
+  models: {
+    providers: {
+      vllm: {
+        baseUrl: "http://127.0.0.1:8000/v1",
+        apiKey: "${VLLM_API_KEY}",
+        api: "openai-completions",
+        models: [
+          {
+            id: "your-model-id",
+            name: "Local vLLM Model",
+            reasoning: false,
+            input: ["text"],
+            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+            contextWindow: 128000,
+            maxTokens: 8192,
+          },
+        ],
+      },
+    },
+  },
+}
+```
+
+## Troubleshooting
+
+- Check the server is reachable:
+
+```bash
+curl http://127.0.0.1:8000/v1/models
+```
+
+- If requests fail with auth errors, set a real `VLLM_API_KEY` that matches your server configuration, or configure the provider explicitly under `models.providers.vllm`.
+
+## Proxy-style behavior
+
+vLLM is treated as a proxy-style OpenAI-compatible `/v1` backend, not a native
+OpenAI endpoint.
+
+- native OpenAI-only request shaping does not apply here
+- no `service_tier`, no Responses `store`, no prompt-cache hints, and no
+  OpenAI reasoning-compat payload shaping
+- hidden Durar attribution headers (`originator`, `version`, `User-Agent`)
+  are not injected on custom vLLM base URLs