nodmix 2026.5.25

package/docs/tools/loop-detection.md

@@ -0,0 +1,154 @@
+---
+summary: "How to enable and tune guardrails that detect repetitive tool-call loops"
+title: "Tool-loop detection"
+read_when:
+  - A user reports agents getting stuck repeating tool calls
+  - You need to tune repetitive-call protection
+  - You are editing agent tool/runtime policies
+  - You hit `compaction_loop_persisted` aborts after a context-overflow retry
+---
+
+Nodmix has two cooperating guardrails for repetitive tool-call patterns:
+
+1. **Loop detection** (`tools.loopDetection.enabled`) — disabled by default. Watches the rolling tool-call history for repeated patterns and unknown-tool retries.
+2. **Post-compaction guard** (`tools.loopDetection.postCompactionGuard`) — enabled by default unless `tools.loopDetection.enabled` is explicitly `false`. Arms after every compaction-retry and aborts the run when the agent emits the same `(tool, args, result)` triple within the window.
+
+Both are configured under the same `tools.loopDetection` block, but the post-compaction guard runs whenever the master switch is not explicitly off. Set `tools.loopDetection.enabled: false` to silence both surfaces.
+
+## Why this exists
+
+- Detect repetitive sequences that do not make progress.
+- Detect high-frequency no-result loops (same tool, same inputs, repeated errors).
+- Detect specific repeated-call patterns for known polling tools.
+- Prevent context-overflow then compaction then same-loop cycles from running indefinitely.
+
+## Configuration block
+
+Global defaults, with every documented field shown:
+
+```json5
+{
+  tools: {
+    loopDetection: {
+      enabled: false, // master switch for the rolling-history detectors
+      historySize: 30,
+      warningThreshold: 10,
+      criticalThreshold: 20,
+      unknownToolThreshold: 10,
+      globalCircuitBreakerThreshold: 30,
+      detectors: {
+        genericRepeat: true,
+        knownPollNoProgress: true,
+        pingPong: true,
+      },
+      postCompactionGuard: {
+        windowSize: 3, // armed after compaction-retry; runs unless enabled is explicitly false
+      },
+    },
+  },
+}
+```
+
+Per-agent override (optional):
+
+```json5
+{
+  agents: {
+    list: [
+      {
+        id: "safe-runner",
+        tools: {
+          loopDetection: {
+            enabled: true,
+            warningThreshold: 8,
+            criticalThreshold: 16,
+          },
+        },
+      },
+    ],
+  },
+}
+```
+
+### Field behavior
+
+| Field                            | Default | Effect                                                                                                                          |
+| -------------------------------- | ------- | ------------------------------------------------------------------------------------------------------------------------------- |
+| `enabled`                        | `false` | Master switch for the rolling-history detectors. Setting `false` also disables the post-compaction guard.                       |
+| `historySize`                    | `30`    | Number of recent tool calls kept for analysis.                                                                                  |
+| `warningThreshold`               | `10`    | Threshold before a pattern is classified as warning-only.                                                                       |
+| `criticalThreshold`              | `20`    | Threshold for blocking repetitive no-progress loop patterns.                                                                    |
+| `unknownToolThreshold`           | `10`    | Block repeated calls to the same unavailable tool after this many misses.                                                       |
+| `globalCircuitBreakerThreshold`  | `30`    | Global no-progress breaker threshold across all detectors.                                                                      |
+| `detectors.genericRepeat`        | `true`  | Warns on repeated same-tool + same-params patterns and blocks when the same calls also return identical outcomes.               |
+| `detectors.knownPollNoProgress`  | `true`  | Detects known polling-like patterns with no state change.                                                                       |
+| `detectors.pingPong`             | `true`  | Detects alternating ping-pong patterns.                                                                                         |
+| `postCompactionGuard.windowSize` | `3`     | Number of post-compaction tool calls during which the guard stays armed and the count of identical triples that aborts the run. |
+
+For `exec`, no-progress checks compare stable command outcomes and ignore volatile runtime metadata such as duration, PID, session ID, and working directory. When a run id is available, recent tool-call history is evaluated only within that run so scheduled heartbeat cycles and fresh runs do not inherit stale loop counts from earlier runs.
+
+## Recommended setup
+
+- For smaller models, set `enabled: true` and leave the thresholds at their defaults. Flagship models rarely need rolling-history detection and can leave the master switch at `false` while still benefiting from the post-compaction guard.
+- Keep thresholds ordered as `warningThreshold < criticalThreshold < globalCircuitBreakerThreshold`.
+- If false positives occur:
+  - Raise `warningThreshold` and/or `criticalThreshold`.
+  - Optionally raise `globalCircuitBreakerThreshold`.
+  - Disable only the specific detector causing issues (`detectors.<name>: false`).
+  - Reduce `historySize` for less strict historical context.
+- To disable everything (including the post-compaction guard), set `tools.loopDetection.enabled: false` explicitly.
+
+## Post-compaction guard
+
+When the runner completes a compaction-retry after a context-overflow, it arms a short-window guard that watches the next few tool calls. If the agent emits the same `(toolName, argsHash, resultHash)` triple multiple times within the window, the guard concludes that compaction did not break the loop and aborts the run with a `compaction_loop_persisted` error.
+
+The guard is gated by the master `tools.loopDetection.enabled` flag with one twist: it stays **enabled when the flag is unset or `true`** and only deactivates when the flag is explicitly `false`. This is intentional. The guard exists to escape compaction loops that would otherwise burn unbounded tokens, so a no-config user still gets the protection.
+
+```json5
+{
+  tools: {
+    loopDetection: {
+      // master switch; set false to disable the guard along with the rolling detectors
+      enabled: true,
+      postCompactionGuard: {
+        windowSize: 3, // default
+      },
+    },
+  },
+}
+```
+
+- Lower `windowSize` is stricter (fewer attempts before abort).
+- Higher `windowSize` gives the agent more recovery attempts.
+- The guard never aborts when results are changing, only when results are byte-identical across the window.
+- It is intentionally narrow: it fires only in the immediate aftermath of a compaction-retry.
+
+<Note>
+  The post-compaction guard runs whenever the master flag is not explicitly `false`, even if you never wrote a `tools.loopDetection` block. To verify, look for `post-compaction guard armed for N attempts` in the gateway log immediately after a compaction event.
+</Note>
+
+## Logs and expected behavior
+
+When a loop is detected, Nodmix reports a loop event and either dampens or blocks the next tool-cycle depending on severity. This protects users from runaway token spend and lockups while preserving normal tool access.
+
+- Warnings come first.
+- Suppression follows when patterns persist past the warning threshold.
+- Critical thresholds block the next tool-cycle and surface a clear loop-detection reason in the run record.
+- The post-compaction guard emits `compaction_loop_persisted` errors with the offending tool name and identical-call count.
+
+## Related
+
+<CardGroup cols={2}>
+  <Card title="Exec approvals" href="/tools/exec-approvals" icon="shield">
+    Allow/deny policy for shell execution.
+  </Card>
+  <Card title="Thinking levels" href="/tools/thinking" icon="brain">
+    Reasoning effort levels and provider-policy interaction.
+  </Card>
+  <Card title="Sub-agents" href="/tools/subagents" icon="users">
+    Spawning isolated agents to bound runaway behavior.
+  </Card>
+  <Card title="Configuration reference" href="/gateway/configuration-reference" icon="gear">
+    Full `tools.loopDetection` schema and merging semantics.
+  </Card>
+</CardGroup>

package/docs/tools/media-overview.md

@@ -0,0 +1,157 @@
+---
+summary: "Image, video, music, speech, and media-understanding capabilities at a glance"
+read_when:
+  - Looking for an overview of Nodmix's media capabilities
+  - Deciding which media provider to configure
+  - Understanding how async media generation works
+title: "Media overview"
+sidebarTitle: "Media overview"
+---
+
+Nodmix generates images, videos, and music, understands inbound media
+(images, audio, video), and speaks replies aloud with text-to-speech. All
+media capabilities are tool-driven: the agent decides when to use them based
+on the conversation, and each tool only appears when at least one backing
+provider is configured.
+
+Live speech uses the Talk session contract instead of the one-shot media tool
+path. Talk has three modes: provider-native `realtime`, local or streaming
+`stt-tts`, and `transcription` for observe-only speech capture. Those modes
+share provider catalogs, event envelopes, and cancellation semantics with
+telephony, meetings, browser realtime, and native push-to-talk clients.
+
+## Capabilities
+
+<CardGroup cols={2}>
+  <Card title="Image generation" href="/tools/image-generation" icon="image">
+    Create and edit images from text prompts or reference images via
+    `image_generate`. Async in chat sessions — runs in the background and
+    posts the result when ready.
+  </Card>
+  <Card title="Video generation" href="/tools/video-generation" icon="video">
+    Text-to-video, image-to-video, and video-to-video via `video_generate`.
+    Async — runs in the background and posts the result when ready.
+  </Card>
+  <Card title="Music generation" href="/tools/music-generation" icon="music">
+    Generate music or audio tracks via `music_generate`. Async in chat
+    sessions on the shared media-generation task lifecycle.
+  </Card>
+  <Card title="Text-to-speech" href="/tools/tts" icon="microphone">
+    Convert outbound replies to spoken audio via the `tts` tool plus
+    `messages.tts` config. Synchronous.
+  </Card>
+  <Card title="Media understanding" href="/nodes/media-understanding" icon="eye">
+    Summarize inbound images, audio, and video using vision-capable model
+    providers and dedicated media-understanding plugins.
+  </Card>
+  <Card title="Speech-to-text" href="/nodes/audio" icon="ear-listen">
+    Transcribe inbound voice messages through batch STT or Voice Call
+    streaming STT providers.
+  </Card>
+</CardGroup>
+
+## Provider capability matrix
+
+| Provider    | Image | Video | Music | TTS | STT | Realtime voice | Media understanding |
+| ----------- | :---: | :---: | :---: | :-: | :-: | :------------: | :-----------------: |
+| Alibaba     |       |   ✓   |       |     |     |                |                     |
+| BytePlus    |       |   ✓   |       |     |     |                |                     |
+| ComfyUI     |   ✓   |   ✓   |   ✓   |     |     |                |                     |
+| DeepInfra   |   ✓   |   ✓   |       |  ✓  |  ✓  |                |          ✓          |
+| Deepgram    |       |       |       |     |  ✓  |       ✓        |                     |
+| ElevenLabs  |       |       |       |  ✓  |  ✓  |                |                     |
+| fal         |   ✓   |   ✓   |   ✓   |     |     |                |                     |
+| Google      |   ✓   |   ✓   |   ✓   |  ✓  |     |       ✓        |          ✓          |
+| Gradium     |       |       |       |  ✓  |     |                |                     |
+| Local CLI   |       |       |       |  ✓  |     |                |                     |
+| Microsoft   |       |       |       |  ✓  |     |                |                     |
+| MiniMax     |   ✓   |   ✓   |   ✓   |  ✓  |     |                |                     |
+| Mistral     |       |       |       |     |  ✓  |                |                     |
+| OpenAI      |   ✓   |   ✓   |       |  ✓  |  ✓  |       ✓        |          ✓          |
+| OpenRouter  |   ✓   |   ✓   |   ✓   |  ✓  |  ✓  |                |          ✓          |
+| Qwen        |       |   ✓   |       |     |     |                |                     |
+| Runway      |       |   ✓   |       |     |     |                |                     |
+| SenseAudio  |       |       |       |     |  ✓  |                |                     |
+| Together    |       |   ✓   |       |     |     |                |                     |
+| Vydra       |   ✓   |   ✓   |       |  ✓  |     |                |                     |
+| xAI         |   ✓   |   ✓   |       |  ✓  |  ✓  |                |          ✓          |
+| Xiaomi MiMo |   ✓   |       |       |  ✓  |     |                |          ✓          |
+
+<Note>
+Media understanding uses any vision-capable or audio-capable model registered
+in your provider config. The matrix above lists providers with dedicated
+media-understanding support; most multimodal LLM providers (Anthropic, Google,
+OpenAI, etc.) can also understand inbound media when configured as the active
+reply model.
+</Note>
+
+## Async vs synchronous
+
+| Capability     | Mode         | Why                                                                                                  |
+| -------------- | ------------ | ---------------------------------------------------------------------------------------------------- |
+| Image          | Asynchronous | Provider processing can outlive a chat turn; generated attachments use the shared completion path.   |
+| Text-to-speech | Synchronous  | Provider responses return in seconds; attached to the reply audio.                                   |
+| Video          | Asynchronous | Provider processing takes 30 s to several minutes; slow queues can run up to the configured timeout. |
+| Music          | Asynchronous | Same provider-processing characteristic as video.                                                    |
+
+For async tools, Nodmix submits the request to the provider, returns a task
+id immediately, and tracks the job in the task ledger. The agent continues
+responding to other messages while the job runs. When the provider finishes,
+Nodmix wakes the agent with the generated media paths so it can tell the
+user and relay the result through the message tool. Nodmix treats missing
+message-tool delivery evidence as a failed completion attempt and does not
+auto-post the generated media as a fallback.
+
+## Speech-to-text and Voice Call
+
+Deepgram, DeepInfra, ElevenLabs, Mistral, OpenAI, OpenRouter, SenseAudio, and xAI can all transcribe
+inbound audio through the batch `tools.media.audio` path when configured.
+Channel plugins that preflight a voice note for mention gating or command
+parsing mark the transcribed attachment on the inbound context, so the shared
+media-understanding pass reuses that transcript instead of making a second
+STT call for the same audio.
+
+Deepgram, ElevenLabs, Mistral, OpenAI, and xAI also register Voice Call
+streaming STT providers, so live phone audio can be forwarded to the selected
+vendor without waiting for a completed recording.
+
+For live user conversations, prefer [Talk mode](/nodes/talk). Batch audio
+attachments stay on the media path; browser realtime, native push-to-talk,
+telephony, and meeting audio should use Talk events and the session-scoped
+catalogs returned by the Gateway.
+
+## Provider mappings (how vendors split across surfaces)
+
+<AccordionGroup>
+  <Accordion title="Google">
+    Image, video, music, batch TTS, backend realtime voice, and
+    media-understanding surfaces.
+  </Accordion>
+  <Accordion title="OpenAI">
+    Image, video, batch TTS, batch STT, Voice Call streaming STT, backend
+    realtime voice, and memory-embedding surfaces.
+  </Accordion>
+  <Accordion title="DeepInfra">
+    Chat/model routing, image generation/editing, text-to-video, batch TTS,
+    batch STT, image media understanding, and memory-embedding surfaces.
+    DeepInfra-native rerank/classification/object-detection models are not
+    registered until Nodmix has dedicated provider contracts for those
+    categories.
+  </Accordion>
+  <Accordion title="xAI">
+    Image, video, search, code-execution, batch TTS, batch STT, and Voice
+    Call streaming STT. xAI Realtime voice is an upstream capability but is
+    not registered in Nodmix until the shared realtime-voice contract can
+    represent it.
+  </Accordion>
+</AccordionGroup>
+
+## Related
+
+- [Image generation](/tools/image-generation)
+- [Video generation](/tools/video-generation)
+- [Music generation](/tools/music-generation)
+- [Text-to-speech](/tools/tts)
+- [Media understanding](/nodes/media-understanding)
+- [Audio nodes](/nodes/audio)
+- [Talk mode](/nodes/talk)

package/docs/tools/minimax-search.md

@@ -0,0 +1,102 @@
+---
+summary: "MiniMax Search via the Token Plan search API"
+read_when:
+  - You want to use MiniMax for web_search
+  - You need a MiniMax Token Plan key or OAuth token
+  - You want MiniMax CN/global search host guidance
+title: "MiniMax search"
+---
+
+Nodmix supports MiniMax as a `web_search` provider through the MiniMax
+Token Plan search API. It returns structured search results with titles, URLs,
+snippets, and related queries.
+
+## Get a Token Plan credential
+
+<Steps>
+  <Step title="Create a key">
+    Create or copy a MiniMax Token Plan key from
+    [MiniMax Platform](https://platform.minimax.io/user-center/basic-information/interface-key).
+    OAuth setups can reuse `MINIMAX_OAUTH_TOKEN` instead.
+  </Step>
+  <Step title="Store the key">
+    Set `MINIMAX_CODE_PLAN_KEY` in the Gateway environment, or configure via:
+
+    ```bash
+    nodmix configure --section web
+    ```
+
+  </Step>
+</Steps>
+
+Nodmix also accepts `MINIMAX_CODING_API_KEY`, `MINIMAX_OAUTH_TOKEN`, and
+`MINIMAX_API_KEY` as env aliases. `MINIMAX_API_KEY` should point at a
+search-enabled Token Plan credential; ordinary MiniMax model API keys may not
+be accepted by the Token Plan search endpoint.
+
+## Config
+
+```json5
+{
+  plugins: {
+    entries: {
+      minimax: {
+        config: {
+          webSearch: {
+            apiKey: "sk-cp-...", // optional if a MiniMax Token Plan env var is set
+            region: "global", // or "cn"
+          },
+        },
+      },
+    },
+  },
+  tools: {
+    web: {
+      search: {
+        provider: "minimax",
+      },
+    },
+  },
+}
+```
+
+**Environment alternative:** set `MINIMAX_CODE_PLAN_KEY`, `MINIMAX_CODING_API_KEY`,
+`MINIMAX_OAUTH_TOKEN`, or `MINIMAX_API_KEY` in the Gateway environment.
+For a gateway install, put it in `~/.nodmix/.env`.
+
+## Region selection
+
+MiniMax Search uses these endpoints:
+
+- Global: `https://api.minimax.io/v1/coding_plan/search`
+- CN: `https://api.minimaxi.com/v1/coding_plan/search`
+
+If `plugins.entries.minimax.config.webSearch.region` is unset, Nodmix resolves
+the region in this order:
+
+1. `tools.web.search.minimax.region` / plugin-owned `webSearch.region`
+2. `MINIMAX_API_HOST`
+3. `models.providers.minimax.baseUrl`
+4. `models.providers.minimax-portal.baseUrl`
+
+That means CN onboarding or `MINIMAX_API_HOST=https://api.minimaxi.com/...`
+automatically keeps MiniMax Search on the CN host too.
+
+Even when you authenticated MiniMax through the OAuth `minimax-portal` path,
+web search still registers as provider id `minimax`; the OAuth provider base URL
+is used as a region hint for CN/global host selection, and `MINIMAX_OAUTH_TOKEN`
+can satisfy the MiniMax Search bearer credential.
+
+## Supported parameters
+
+| Parameter | Type    | Constraints | Description                                                                 |
+| --------- | ------- | ----------- | --------------------------------------------------------------------------- |
+| `query`   | string  | required    | Search query string.                                                        |
+| `count`   | integer | 1-10        | Number of results to return. Nodmix trims the returned list to this size. |
+
+Provider-specific filters are not currently supported.
+
+## Related
+
+- [Web Search overview](/tools/web) -- all providers and auto-detection
+- [MiniMax](/providers/minimax) -- model, image, speech, and auth setup