Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.
Python agentagent-osaiai-agentsai-partnerchatopsclaudeclaude-codecodexdevtoolsdiscord-botlark-botllmlocal-firstlong-term-memoryopencodeslack-botvibe-codingvibeywechat
feat(model-hub): turn on web_fetch (WF-3) (#2982)
Call White08adc50687f5
diff --git a/config/v2_config.py b/config/v2_config.pyindex 6871229b1..f49da2973 100644--- a/config/v2_config.py+++ b/config/v2_config.py@@ -3045,8 +3045,8 @@ class ModelHubAgentSupplyConfig:# through the executor, and the direct image tool.MODEL_HUB_SEARCH_TOOLS: tuple[str, ...] = ("web_search", "x_search")MODEL_HUB_HOSTED_TOOLS: tuple[str, ...] = (*MODEL_HUB_SEARCH_TOOLS, "image_generate", "web_fetch")-# Readers precede every producer release; activation is coordinated with public disclosure.-WEB_FETCH_ACTIVE = False+# WF-3 enables read, declaration and producer together; coordinated release decision: docs/plans/web-fetch.md §8.+WEB_FETCH_ACTIVE = Truedef _drop_vibey_client_presets(agent: "ModelHubAgentSupplyConfig") -> None:diff --git a/core/agent_core/agent/loop.py b/core/agent_core/agent/loop.pyindex c1aa5f1cf..64b91ba8b 100644--- a/core/agent_core/agent/loop.py+++ b/core/agent_core/agent/loop.py@@ -152,6 +152,7 @@ from core.agent_core.hosted_continuations import pending_hostedfrom core.agent_core.messages import (AssistantMessage,HostedImage,+ HostedRecord,Message,Origin,ToolCallBlock,@@ -1433,7 +1434,8 @@ class Agent:self._rows.append(row)# The committed result, which a settlement race may have taken from another writer, not this execution's.committed = row.message.hosted if isinstance(row.message, ToolResultMessage) else None- if committed is not None and committed.record.status == "completed" and call.name == "image_generate":+ if (committed is not None and isinstance(committed.record, HostedRecord)+ and committed.record.status == "completed" and call.name == "image_generate"):# Only an image a surface can deliver counts toward the reply, and is carried with it.self._generated_images.extend(image for image in committed.record.images or () if image_deliverable(image))if emit_finished:diff --git a/core/handlers/model_hub/service.py b/core/handlers/model_hub/service.pyindex 47e3f8962..dbfafe9c8 100644--- a/core/handlers/model_hub/service.py+++ b/core/handlers/model_hub/service.py@@ -7305,7 +7305,7 @@ class ModelHubService:return {}async def _run_fetch_chain(self, call: ExecutorCall, turn: TurnHostedTools, *, say: Callable[..., str]) -> dict:- """Run the dormant web_fetch service/local chain behind the activation flag."""+ """Run the web_fetch service/local chain behind the shared activation switch."""from .fetch_chain import FetchChainreturn await FetchChain(self, call, turn, say).run()diff --git a/docs/plans/agent-core-contracts/README.md b/docs/plans/agent-core-contracts/README.mdindex 63cacba3c..623bd3d69 100644--- a/docs/plans/agent-core-contracts/README.md+++ b/docs/plans/agent-core-contracts/README.md@@ -28,8 +28,10 @@ Conventions:are unchanged. Executor origins accept `fetch_api` and `local_fetch`, notassistant-model origins. Attempts accept `local`.- C-6 recognizes function-only fetch facts (`native` and `replay` both- `unsupported`). The dormant build never declares or executes fetch; records,- activity and usage remain readable. #2920's released rc4 reader preserves the+ `unsupported`). WF-3 enables declaration and execution through the shared+ `WEB_FETCH_ACTIVE` switch; records, activity and usage remain readable even+ when the switch is false. The coordinated reader/activation release decision is+ in `../web-fetch.md` §8. #2920's released rc4 reader preserves thefull ordinary quotation when discarding unknown hosted metadata. Config andusage downgrade writes are intentionally lossy as disclosed in the owning plan.diff --git a/docs/plans/agent-core-contracts/model-hub-consumer.md b/docs/plans/agent-core-contracts/model-hub-consumer.mdindex 06ae21926..415c391d0 100644--- a/docs/plans/agent-core-contracts/model-hub-consumer.md+++ b/docs/plans/agent-core-contracts/model-hub-consumer.md@@ -68,17 +68,18 @@ consumer's requirements, and the Model Hub lane lands the change in both places.7. **Hosted tools** (v2, [`../vibey-hosted-tools.md`](../vibey-hosted-tools.md) §3.4, §3.5, §6, §7).- `HopResolution.hosted_tools` has one entry per hosted tool the turn declares:- - **`web_fetch`** (WF-1/WF-2V, unreleased v21) has exactly+ - **`web_fetch`** (WF-1/2/3, v21) has exactly`{native: "unsupported", replay: "unsupported"}`: always function-only,- never Claude-native. Both its server entry and Vibey registration remain- absent while `WEB_FETCH_ACTIVE = False`. When active, the first resolution+ never Claude-native. WF-3 enables its server entry, Vibey registration and+ producer with `WEB_FETCH_ACTIVE = True`; an explicit false hides them and+ rejects execution. The first resolutionfreezes its canonical `{url}`-only function beside the other turn tools.Vibey posts `{"arguments": {"url": ...}}` under the live turn to`/hosted-tools/web_fetch`; the gateway owns normalization and the budget.The validated record produces C-1's complete capped ordinary quotationthrough the same deterministic frame used by readers. Replay preserves- the stored call/result without a fetch or settlement. Activation belongs- to WF-3, after a reader release.+ the stored call/result without a fetch or settlement. The owner chose to+ ship readers and activation together (`../web-fetch.md` §8).- **a search tool** (`web_search`, `x_search`) gives the primary hop's facts:- `native` is `assumed` when the tool runs on the hop (`hosted_tools.runs_on`: a hub Source with an eligiblecredential, never a Claude subscription, and an untranslated path, that serves the model, for which thediff --git a/docs/plans/model-hub-contracts/README.md b/docs/plans/model-hub-contracts/README.mdindex 108502100..1fd601922 100644--- a/docs/plans/model-hub-contracts/README.md+++ b/docs/plans/model-hub-contracts/README.md@@ -1,7 +1,7 @@# Model Hub contracts-Status: **contract_version 21**, 2026-10-25: keyless Exa activation (KEX-3)-and dormant fetch readers (WF-1) / Vibey and UI producer support (WF-2V).+Status: **contract_version 21**, 2026-10-26: keyless Exa activation (KEX-3)+and web fetch readers, producers and coordinated activation (WF-1/2/3).`gh-v3.2.3rc3` through `rc7` carry v20, so the unreleased v21 event now includesits coordinated readers, producer, and default-on Exa search activation. The versionedmirrors advance together; persisted TurnProvenance retains v5–v20 and adds v21.@@ -16,8 +16,9 @@ WF-1 appends dormant `web_fetch` readers to this unreleased event:separate fetch consent/order, local/Exa/Jina synthetic identities, fetch/localexecutor origins and local attempts, C-1 bounded fetch records, C-6 function-onlyfacts, and optional service usage `tool` attribution (absent means `web_search`).-`WEB_FETCH_ACTIVE = False` keeps GET/candidates/declarations/active settings UI-unchanged; historical reads and V2/PUT preservation do not depend on activation.+WF-1 originally kept `WEB_FETCH_ACTIVE = False`, leaving GET/candidates/+declarations/active settings UI unchanged; historical reads and V2/PUT preservation+do not depend on activation.The exact shapes and downgrade losses are owned by [`../web-fetch.md`](../web-fetch.md)§§2/4/6/8. WF-1 adds no fetch producer, network transport or dependency.The localized hosted-activity vocabulary adds `service_rate_limited` as a fallback@@ -33,7 +34,18 @@ activity projection adds optional `policy_sources` (service ids reached beforean address-policy refusal; an empty list means not forwarded). Web and IM readthat one decision, localize synthetic labels, and retain historical readabilityregardless of activation. See [`../web-fetch-vibey-ui.md`](../web-fetch-vibey-ui.md)-for fixture acceptance and intentional boundaries. The production flag stays off.+for fixture acceptance and intentional boundaries.++WF-3 sets `WEB_FETCH_ACTIVE = True` across the read, candidates, frozen declaration+and gateway producer after the WF-2G engine/chain/provider PRs (#2962/#2968/#2976).+It changes no serialized shape or contract version. Missing fetch config uses the+automatic local → configured services → keyless Exa → keyless Jina order without+writing defaults; saved fetch opt-outs and ordering remain independent of search.+Public en/zh disclosure ships with activation (`avibe-docs#83`); §7 of the owning+plan specifies up/down buttons, not drag. **Owner decision, 2026-10-26 01:56+Asia/Shanghai:** ship WF-1 readers and the activation flip together; no separate+prior reader release is required. Exact-head review/CI and coordinated activation/+disclosure acceptance still apply. See [`../web-fetch.md`](../web-fetch.md) §8.It builds on v20 (2026-10-22): active web-search services (S2b-2).`gh-v3.2.3rc1` and `gh-v3.2.3rc2` carried v19, so activation opened v20 rather thandiff --git a/docs/plans/model-hub-contracts/api.md b/docs/plans/model-hub-contracts/api.mdindex 6ac87c849..2a35cf5fc 100644--- a/docs/plans/model-hub-contracts/api.md+++ b/docs/plans/model-hub-contracts/api.md@@ -73,7 +73,7 @@ runtime open. Once models are configured it participates in the runtime Stop gua| GET `/api/models/quota` | → `{quota: QuotaSummary}` | Rate-limit windows of every hub-held subscription Source, from the vendor's own usage report, plus each Source's API-price `value` (see Subscription quota). Served from the server cache; a Source older than five minutes is re-read before answering, bounded by a short deadline, and a Source still being read is listed in `pending`. Never starts the engine. || POST `/api/models/quota/refresh` | → `{quota: QuotaSummary}` | Forced re-read. Rate-limited per Source to one vendor call every 30 seconds and suppressed while a Source cools down after a vendor 429; a suppressed Source returns its cached snapshot. || GET `/api/models/hosted-tools` | → `{hosted_tools: HostedToolsStatus}` | Models → Tools (`hosted-tools-status.schema.json`): every catalog tool with its switch (absent setting = on), its passive status from real use (`latest_ok`, current `refused[]`, per-pair `verified[]`), for a search tool the saved and effective executor order with every eligible executor candidate and its unit rates, and for `image_generate` the saved and effective default and the available image models. Memory only; nothing is probed. |-| POST `/api/models/hosted-tools/web_fetch/services/<source_id>/retry` | `{}` → `{}` | WF-1 reader/control only: clears the same Source/service cooldown and generation as search for Exa (`src_exakeyless`, `exa/mcp`), or Jina's own. Member management role; no fetch is sent. Local has no whole-executor retry and returns `404 source_not_found`. |+| POST `/api/models/hosted-tools/web_fetch/services/<source_id>/retry` | `{}` → `{}` | Clears the same Source/service cooldown and generation as search for Exa (`src_exakeyless`, `exa/mcp`), or Jina's own. Member management role; no fetch is sent. Local has no whole-executor retry and returns `404 source_not_found`. || POST `/api/models/hosted-tools/web_search/services/<source_id>/retry` | `{}` → `{}` | Clears the controller's `(Source, service)` cooldown, advances its generation, and leaves Source health unchanged. The Tools read is the follow-up source of truth. Member management role. || PUT `/api/models/hosted-tools/<tool>` | `{enabled?: boolean, executor_order?: null \| [source_id], default_model?: null \| {source_id, model_id}, prefer_native?: boolean, services?: {source_id: null \| {results?: integer \| null, enabled?: boolean \| null}}}` → `{hosted_tools: HostedToolsStatus}` | Changes one tool's settings field by field: an absent field is unchanged, `null` resets `executor_order` to automatic or `default_model` to the default. `executor_order` is for search/fetch tools and `default_model` for `image_generate`; values are saved as given, never resolved. It applies from the next turn and interrupts nothing. An unknown tool returns `404 hosted_tool_not_found`; an empty or malformed body `400 invalid_parameter`, including a null `enabled`, an id that is not a Source id (`src_…`), or a model id with surrounding whitespace. || POST `/api/models/hosted-tools/<tool>/clear` | `{source_id?, model_id?}` → `{hosted_tools: HostedToolsStatus}` | Clears the tool's refusals (`call` and `replay`): on one Source and model, on one Source (every model), or with `{}` on every Source. `model_id` requires `source_id`; neither is nullable, so a malformed body is refused rather than widened into a broader clear. A Source or model with nothing to clear is a no-op that answers the read. The file is written first: a failed write returns `500 state_write_failed` and changes nothing. A cleared Source is assumed again from its next request. An unknown tool returns `404 hosted_tool_not_found`; any other body `400 invalid_parameter`. |@@ -140,7 +140,7 @@ reader and producer together, with the public disclosure shipped in the same v21event.PUT accepts `prefer_native` only for `web_search`; search's `services` patch includes-`results`, while fetch accepts enabled-only patches (see WF-1 below). `prefer_native`+`results`, while fetch accepts enabled-only patches (see Web fetch below). `prefer_native`is a boolean, not nullable. `services` is a Source-keyed patch: an absent Sourceis untouched, a present object merges only its supplied fields, and a nullSource or null field removes its saved override. An empty patch is a no-op.@@ -180,12 +180,15 @@ orders. Re-upgrading then uses service defaults; reapply DeepSeek's opt-outbefore enabling service execution. This is not a promise to retain new choicesthrough an older release's write.-### Dormant fetch readers (WF-1)+### Web fetch (WF-1 readers, WF-3 activation)-`web_fetch` is a known saved tool in contract v21, but while `WEB_FETCH_ACTIVE`-is false it is absent from GET, effective candidates, model declarations and-active settings UI. V2 and PUT retain its settings; historical activity and-usage remain readable. There is no fetch execution endpoint in WF-1.+`web_fetch` is active in contract v21. The shared `WEB_FETCH_ACTIVE = True`+switch enables GET, effective candidates, model declarations and the gateway+producer together. An explicit false still omits the active surfaces and refuses+execution; V2/PUT settings and historical activity/usage remain readable.+The canonical `{url}` function uses `/hosted-tools/web_fetch` on the live Vibey+turn; it is never a native Claude fetch. The owner's coordinated reader/activation+release decision and public disclosure are in `web-fetch.md` §8.PUT accepts `{enabled?, executor_order?, services?}` for fetch: each service patchhas only nullable `enabled`; no `results`, `prefer_native` or client-written@@ -193,7 +196,7 @@ roles. Null resets an override/order, absent fields are untouched. An explicitorder save captures the current `local`/`service` roles. Independent ordering,automatic insertion and disabled-tail rules are specified in `web-fetch.md` §2.-The future active GET entry is `FetchTool` in `hosted-tools-status.schema.json`:+The active GET entry is `FetchTool` in `hosted-tools-status.schema.json`:`services` contains saved partial consent; candidate `settings` is `{enabled}`.`service_vendors` carries `capabilities: ["web_fetch"]`, `content_types`, price,Source ids and `{built_in, requires_setup, auto_enroll}`. The local executor anddiff --git a/docs/plans/model-hub-contracts/mirror-registry.json b/docs/plans/model-hub-contracts/mirror-registry.jsonindex 7906f1a73..9cab134cb 100644--- a/docs/plans/model-hub-contracts/mirror-registry.json+++ b/docs/plans/model-hub-contracts/mirror-registry.json@@ -425,7 +425,7 @@"decision_tables": [{"id": "D31",- "domain": "readable hosted result tools including dormant fetch",+ "domain": "readable hosted result tools including fetch","authority": {"kind": "schema_vocabulary", "schema": "capability-evidence.schema.json", "paths": ["/properties/feature/oneOf/1/properties/name"]},"consumers": [{"kind": "typescript_string_union", "file": "ui/src/lib/hostedActivity.ts", "name": "HostedToolName"}]},diff --git a/docs/plans/web-fetch-gateway-delivery.md b/docs/plans/web-fetch-gateway-delivery.mdindex 38369ad61..1c49ae970 100644--- a/docs/plans/web-fetch-gateway-delivery.md+++ b/docs/plans/web-fetch-gateway-delivery.md@@ -1,24 +1,27 @@# WF-2G gateway delivery-Status: 2026-10-25, dormant implementation; `WEB_FETCH_ACTIVE=False`.+Status: 2026-10-26, WF-2G complete: #2962, #2968 and #2976 merged.+WF-2G kept `WEB_FETCH_ACTIVE=False`; WF-3 now prepares the coordinated activation+and public disclosure. The owner chose a combined reader/activation release+(see §8 of the owning plan). The sections below retain the split's history.The behavior contract remains [web-fetch.md](web-fetch.md), especially §§3–6.This plan narrows delivery units, not the activation/release scope.## Delivery units and ownership1. **Local engine and packaging:** PR #2962 merged. No activation.-2. **Chain and keyless services:** PR #2968 retains the URL/DNS gate, local+2. **Chain and keyless services:** PR #2968 merged with the URL/DNS gate, localwiring, keyless Exa/Jina, ordered fallback, no-send/skipped contract,additive fetch usage attribution, and endpoint routing.-3. **Configured services:** a separate PR from the updated `master` after- #2968 merges adds Tavily Extract, keyed Exa `/contents`, and DeepSeek Fetch.- Both producer PRs must merge before WF-3 enables the feature.+3. **Configured services:** PR #2976 merged from updated `master` after #2968,+ adding Tavily Extract, keyed Exa `/contents`, and DeepSeek Fetch.+ Both producer prerequisites are complete; WF-3 still requires review/CI and acceptance.The gateway lane prepares exact-head review/CI evidence; the orchestrator ownsmerging. No service restart, regression deployment, release or live paid probe ispart of this lane. Tests use test-owned state and loopback fixtures only.-## Convergence boundary+## Convergence boundary (historical #2968 split)The third reviewed head of #2968 again had multiple provider-mapping/settlementfindings, triggering the orchestrator's split guard. The configured transport,@@ -40,7 +43,7 @@ response → settlement lifecycle. Boundary tests now own these invariants:| Boundary | Required evidence || --- | --- |-| Dormancy | Global flag rejects stale turns; configured providers cannot reach either adapter or transport even in flag-enabled test fixtures. |+| Activation boundary | Global flag rejects stale turns before local or configured transport admission. Configured providers became eligible only with #2976; WF-3 exercises defaults without flag-on test patches. || Admission | Every service submission re-gates only the original URL; a refusal stops forwarding but retains usage for earlier sent attempts. || Attempt ownership | One capped deadline starts before the shared Exa lock, includes subsequent transport work, and allows Jina fallback on expiry. || No-send | Disabled/unavailable/contended/over-budget candidates have only skipped facts, no usage, no capability success and no cooldown write. |@@ -56,10 +59,10 @@ fixture in `tests/test_model_hub_keyless_exa.py`. Existing search-chain and fetcreader tests cover shared cooldown/ledger/read compatibility. Local engine testsremain the gate/transport/extraction owner; no live providers are test or CI inputs.-## Configured-provider follow-up acceptance+## Configured-provider follow-up acceptance (delivered in #2976)-Build one explicit table across provider × response phase × failure scope before-implementing the smaller PR:+The separate follow-up used an explicit table across provider × response phase ×+failure scope:- request shapes: Tavily basic/Markdown `urls`, Exa `/contents` `ids` and boundedtext, DeepSeek `url`; use only the catalog endpoint and opaque credential owner;diff --git a/docs/plans/web-fetch-vibey-ui.md b/docs/plans/web-fetch-vibey-ui.mdindex c15e4d7dc..da103f5eb 100644--- a/docs/plans/web-fetch-vibey-ui.md+++ b/docs/plans/web-fetch-vibey-ui.md@@ -5,6 +5,12 @@ Change contract, 2026-10-25. Implements the Vibey and UI portion ofowns gateway policy, network transports, extraction, normalization, settlementand packaging. No gateway contract change is needed by this lane.+Activation follow-up, 2026-10-26: WF-3 sets the shared flag to true after all+WF-2G producers merged, with default-on boundary acceptance and public en/zh+disclosure. At 01:56 Asia/Shanghai the owner chose to ship readers and activation+together, without a separate reader-first release. The invariants below describe+WF-2V's original dormant landing; the current contract is `web-fetch.md` §§7–8.+## Invariants- The production `WEB_FETCH_ACTIVE` remains false. No function registration or@@ -53,10 +59,39 @@ The existing WF-1 downgrade tests remain the authority for released-shapeloading and lossy old config/usage rewrites. They are not lossless downgradeclaims. The ordinary text path remains compatible with #2920.+## WF-3 runtime acceptance and hosted-record audit++The activation's direct tool test did not reach transcript commit. Real Vibey+acceptance now exercises Anthropic wire → Agent loop → hosted tool/gateway HTTP+→ result frame → SQLite transcript/media registration → dispatcher/mirror →+activity API, then creates a fresh adapter for next-turn replay with fetch no+longer declared. The table covers Local, keyless Exa, Jina and policy refusal on+Web and Telegram, without live providers. A 200 must produce one committed+`WebFetchRecord`, one card/IM line and an ordinary final answer, not an interrupted+call; replay preserves that row and frame without another fetch or media entry.++Audit of production search/image-only record access:++| Owner | Type boundary |+| --- | --- |+| `storage/agent_transcript.py::_register_images` | Narrow to `HostedRecord` before enumerating images. `WebFetchRecord` is not a media container; previously this generic consumer raised before the result transaction committed. |+| `core/agent_core/agent/loop.py::_commit_result` | Image collection requires both `HostedRecord` and `image_generate`; a fetch can never enter it. |+| `core/agent_core/messages.py::_record_to_dict` | Dispatch fetch serialization before reading search sources/image fields. |+| `core/agent_core/ai/hosted.py::record_text` | Dispatch to the bounded fetch frame before search/image formatting. |+| `core/hosted_activity.py::proxy_activity` | Narrow fetch by type and project empty search sources/images plus the fetch record. Native activities accept only `HostedToolBlock`, whose record is validated as `HostedRecord`. |+| Vibey image tool, media sink and shared mirror | Image tool decoding validates the image tool's record; the media sink accepts image bytes, not generic records. The mirror reads the projected card's image list, empty for fetch. |++Primary regression: `tests/test_vibey_agent.py::test_web_fetch_survives_the_real_agent_commit_delivery_and_next_turn_replay`.+It reproduced the exact `WebFetchRecord` missing `images` failure before the fix.+Existing transcript media-sweep and image-generation tests retain positive image+registration/delivery coverage; no dummy image/search fields are added to fetch.+The orchestrator owns the separate local-Incus live-provider rerun before merge.+## Known by design- No activation, release/tag, public-doc activation, deployment, service restart- or local/remote regression update. WF-3 must first establish a reader release.+ or local/remote regression update in the original WF-2V scope. WF-3's coordinated+ activation and owner-controlled acceptance are described above.- The Tools read has no live local-transport state. Per orchestrator approval,Local detail shows the default Chrome 150 profile, fallback policy and limits;only actual call records show transport degradation and its reason.diff --git a/docs/plans/web-fetch.md b/docs/plans/web-fetch.mdindex 8cadd974c..ff099e133 100644--- a/docs/plans/web-fetch.md+++ b/docs/plans/web-fetch.md@@ -136,13 +136,14 @@ readers skip residual unknown ids without a write.### 2.3 Read and controls-`GET /api/models/hosted-tools` adds `web_fetch` **only after activation**. WF-1-sets `WEB_FETCH_ACTIVE = False`: GET has no fetch entry, candidates, declaration-or active settings UI. V2 and PUT still preserve saved fetch choices, while-historical records, usage and activity are readable regardless of the flag.-This dormant-read boundary was approved on 2026-10-25.--Once active, the entry contains saved `services` and+`GET /api/models/hosted-tools` includes `web_fetch` in WF-3, with the shared+`WEB_FETCH_ACTIVE = True` switch enabling the read, candidates, declaration and+producer together. WF-1/WF-2 kept that switch false until activation; an explicit+false still hides the row/candidates/declaration and rejects producer calls.+V2 and PUT preserve saved fetch choices, while historical records, usage and+activity remain readable regardless of the flag.++The active entry contains saved `services` and`executor_order`, resolved `effective_order`, `executor_candidates`, status andcapability-bearing `service_vendors`. The candidate uses existing service fields:`source_id`, `model_id` (service id), `kind`, `billing`, `relay`, dated `price`,@@ -629,8 +630,10 @@ Built-ins do not become “removed Source” rows without a credential row.Add **网页读取 / Web fetch** after Web search in Models → Tools. Header switch,latest real result and **always-visible** numbered fetch-order list (including-automatic state). Each row has the existing up/down/drag controls, enabled switch,+automatic state). Each row has the existing up/down **buttons only**, enabled switch,kind pill, provider/Local label, price/unit, capability badges and current status.+There is no drag interaction: fetch follows web search's buttons-only ordering+(owner-approved clarification, 2026-10-25).Disabled rows remain at the tail. Reset to automatic writes null, never effectivedefaults. No Claude-native/prefer-native row or “unverified” state.@@ -691,9 +694,53 @@ Python/TypeScript mirrors, public routes and contract registry on one tested hea| Phase | Scope / gate || --- | --- |-| WF-1 readers | Config/read/record/usage/activity readers, synthetic identities, versioned mirrors and downgrade fixtures. Include #2920, release before any fetch producer writes. Hide active UI/candidates until activation. |+| WF-1 readers | Config/read/record/usage/activity readers, synthetic identities, versioned mirrors and downgrade fixtures. Include #2920; originally reader-first, superseded by the coordinated-release decision below. Hide active UI/candidates until activation. || WF-2 producer | Behind `WEB_FETCH_ACTIVE=false`: gateway policy/local transport/extraction, keyed APIs, shared Exa client + Jina, Vibey framing/replay, UI fixtures. Separate lanes only after reader contracts merge. Flag-off preserves prior behavior. |-| WF-3 activation | Confirm reader release exists; flip declaration/candidates/producer together and publish public en/zh docs. Recheck vendor facts/tags, review/CI and manual fixture acceptance. |+| WF-3 activation | Owner-approved coordinated reader/flip release: enable declaration/candidates/producer together and publish public en/zh docs. Recheck vendor facts/tags, review/CI and manual acceptance. |++### WF-3 activation and coordinated release++The activation change sets the one shared switch to true; it changes no schema,+contract version, migration, catalog price or default ordering. It builds on the+merged local engine (#2962), local/keyless chain (#2968), and configured providers+(#2976). Public disclosure is the coordinated `avibe-docs#83` English/Chinese change.++At activation preparation on **2026-10-26**, the latest published release was+`gh-v3.2.3rc7` (v20), and no fetched tag contained WF-1 `9e47e5f`. At **01:56+Asia/Shanghai on 2026-10-26**, the owner explicitly waived the separate reader-first+release and chose to **ship readers and the activation flip together**. No rc8 or+other prior reader release is required. The orchestrator merges the activation+and public-doc PRs together once the activation is Merge-ready and its required+acceptance is complete; exact-head review and CI are not waived. This lane does+not publish a tag/release, deploy or restart a service.++Primary activation acceptance connects the real default-on Tools read, frozen+candidates, C-6 hop, Vibey declaration, gateway chain and bounded C-1 result/usage+without patching either flag. Only external I/O is stubbed. Existing provider,+policy, downgrade and four-protocol replay tests remain in force. Screenshots in+`wf3-shots/` are hermetic real-component/projection fixtures, not live-provider or+installed-runtime verification; §7's buttons match the accepted UI.++### Release note (publish only with activation)++**English:** Web fetch is now available in Vibey and Models → Tools. It reads one+URL through local fetch → configured fetch services → free Exa → free Jina Reader+by default. Fetch switches and up/down ordering are independent of search;+existing eligible API-key Sources are included automatically. Local fetch sends+requests directly from your runtime host without browser credentials or JavaScript+rendering; remote services receive the URL and may render JavaScript or charge+per request. Review the Web fetch documentation for costs, limits and terms.+Stored results are bounded to 48,000 UTF-8 bytes including quotation framing and+replay without fetching or billing again. Older-version config/usage rewrites can+lose fetch opt-outs and attribution; recheck settings after a downgrade/re-upgrade.++**中文:** Vibey 和「模型 → 工具」现已支持网页读取。默认依次尝试本机、已配置的+读取服务、免费 Exa 和免费 Jina Reader;读取开关及上下按钮排序与搜索独立,现有+符合条件的 API 密钥来源会自动加入。本机从运行 Avibe 的主机直接发起请求,不携带+浏览器凭据、不渲染 JavaScript;远端服务会接收网址,可能渲染 JavaScript 或按次收费。+费用、限制及条款请查看网页读取文档。保存结果连同引用框架最多 48,000 UTF-8 字节,+重放不会再次读取或计费。旧版本重写配置或用量可能丢失读取禁用选项及用量归属;+降级后重新升级时请检查设置。Unknown-tool compatibility is #2920's exact ordinary-content degradation, not apromise that old readers understand new record fields or preserve new configdiff --git a/storage/agent_transcript.py b/storage/agent_transcript.pyindex 5be79d1fa..1bb2c6ba9 100644--- a/storage/agent_transcript.py+++ b/storage/agent_transcript.py@@ -77,6 +77,7 @@ from core.agent_core.harness.store import ContextEntry, EntryKindfrom core.agent_core.messages import (AssistantMessage,HostedImage,+ HostedRecord,HostedToolBlock,TextBlock,ToolResultMessage,@@ -467,10 +468,15 @@ def _register_images() -> None:"""Register each generated image ``message`` names, in its row's transaction (module docstring)."""+ # WebFetchRecord shares the hosted envelope, not search/image fields.+ # Narrow before entering media registration so page results can commit.+ records: list[HostedRecord]if isinstance(message, AssistantMessage):records = [block.record for block in message.content if isinstance(block, HostedToolBlock)]+ elif message.hosted is not None and isinstance(message.hosted.record, HostedRecord):+ records = [message.hosted.record]else:- records = [message.hosted.record] if message.hosted is not None else []+ returnfor image in (image for record in records for image in record.images or ()):media_service.register(conn,diff --git a/tests/test_model_hub_fetch_chain.py b/tests/test_model_hub_fetch_chain.pyindex c7ab86d48..4a8a0f04b 100644--- a/tests/test_model_hub_fetch_chain.py+++ b/tests/test_model_hub_fetch_chain.py@@ -22,12 +22,6 @@ from vibe.model_hub_runtime.fetch_catalog import fetch_service_catalogfrom core.handlers.model_hub.service import ModelHubError[email protected](autouse=True)-def _reset_fetch_flag(monkeypatch):- monkeypatch.setattr(hosted_tools, "WEB_FETCH_ACTIVE", False)- yield--@pytest.mark.asyncioasync def test_jina_builtin_fetch_is_admitted_with_reader_headers():adapter = object.__new__(CLIProxyEngineAdapter)@@ -94,8 +88,8 @@ def test_http_200_malformed_provider_envelopes_are_service_failures(body, vendor@pytest.mark.asyncio@pytest.mark.parametrize("vendor", ["exa", "tavily", "deepseek"])-async def test_configured_fetch_is_dormant_at_chain_and_adapter_boundaries(tmp_path, vendor):- """The activation flag still blocks the new producer while the adapter is ready for hermetic tests."""+async def test_flag_off_rejects_configured_fetch_from_a_previously_frozen_turn(tmp_path, monkeypatch, vendor):+ """The shared kill switch rejects stale turns before any configured transport admission."""sid = IDS[vendor]service, config = _service(tmp_path, vendor=vendor,@@ -109,7 +103,7 @@ async def test_configured_fetch_is_dormant_at_chain_and_adapter_boundaries(tmp_pservice.adapter.fetch = unexpectedturn = service.freeze_turn_hosted_tools(config)- hosted_tools.WEB_FETCH_ACTIVE = False+ monkeypatch.setattr(hosted_tools, "WEB_FETCH_ACTIVE", False)with pytest.raises(ModelHubError) as raised:await service.run_hosted_tool("web_fetch", {"arguments": {"url": "https://example.com/"}},@@ -626,7 +620,6 @@ def _service(tmp_path: Path, *, enabled: dict[str, bool], order=(LOCAL, EXA), ve)service.store.config = configservice._local_fetch_engine = _Local()- hosted_tools.WEB_FETCH_ACTIVE = Truereturn service, config@@ -634,7 +627,7 @@ def _service(tmp_path: Path, *, enabled: dict[str, bool], order=(LOCAL, EXA), veasync def test_flag_off_rejects_fetch_without_running_gate_or_adapter(tmp_path, monkeypatch):service, config = _service(tmp_path, enabled={LOCAL: True, EXA: True, FREE_EXA: False, JINA: False})turn = service.freeze_turn_hosted_tools(config)- hosted_tools.WEB_FETCH_ACTIVE = False+ monkeypatch.setattr(hosted_tools, "WEB_FETCH_ACTIVE", False)called = []monkeypatch.setattr(service.adapter, "fetch", lambda *args, **kwargs: called.append(args), raising=False)with pytest.raises(ModelHubError) as raised:@@ -644,25 +637,75 @@ async def test_flag_off_rejects_fetch_without_running_gate_or_adapter(tmp_path,)assert raised.value.code == "tool_not_enabled"assert called == []+ assert service._local_fetch_engine.checks == 0@pytest.mark.asyncio-async def test_local_sent_attempt_is_recorded_as_exact_zero_fetch_usage(tmp_path):- service, config = _service(- tmp_path,- enabled={LOCAL: True, EXA: False, FREE_EXA: False, JINA: False},- order=(LOCAL, EXA),- )- result = await service.run_hosted_tool(- "web_fetch", {"arguments": {"url": "https://example.com/"}},- turn=service.freeze_turn_hosted_tools(config),- )+async def test_default_activation_connects_tools_read_frozen_hop_and_bounded_local_result(tmp_path):+ """WF-3: no flag patches; a read/declaration/producer mismatch must break this path.++ Earlier lane fixtures forced activation or supplied a canned gateway answer.+ Only the external local engine and HTTP transport are replaced here.+ """+ import httpx++ from config import v2_config+ from core.agent_core.messages import ToolResultMessage, WebFetchRecord, message_from_dict, message_to_dict+ from modules.agents.model_hub import ModelHubLaunch+ from modules.agents.vibey.models import selection_from_hop+ from modules.agents.vibey.tools import hosted_tools as declared_tools+ from tests.test_vibey_web_fetch import router++ service = owner(tmp_path, (), model=True)+ service._local_fetch_engine = _Local()+ config = service.store.load()+ before = config.to_payload()+ assert "web_fetch" not in config.hosted_tools+ assert v2_config.WEB_FETCH_ACTIVE and hosted_tools.WEB_FETCH_ACTIVE+ fetch = next(row for row in service.hosted_tools_status()["hosted_tools"]["tools"] if row["name"] == "web_fetch")+ assert fetch["enabled"] is True+ assert fetch["effective_order"] == [LOCAL, FREE_EXA, JINA]+ frozen = service.freeze_turn_hosted_tools(config)+ assert [slot.source_id for slot in frozen.executors["web_fetch"]] == [LOCAL, FREE_EXA, JINA]+ source = config.sources[0]+ model = source.models[0].id+ facts = service.hosted_tools_for_hop(source, model, frozen)+ assert facts["web_fetch"] == {"native": "unsupported", "replay": "unsupported"}+ hop = ModelHubLaunch(+ backend="vibey", channel="hub", requested_model=model, target_model=model, runtime_model=model,+ source_id=source.id, source_identity="test-identity", gateway_base_url="http://hub.invalid/vibey",+ gateway_token="fixture", gateway_request_metadata={"x-avibe-turn-id": "turn-fetch"},+ protocol="openai_chat", provider="openai", hosted_tools=facts,+ ).to_hop_resolution()+ sent = []++ async def gateway(request):+ assert request.url.path == "/vibey/v1/hosted-tools/web_fetch"+ assert request.headers["x-avibe-turn-id"] == "turn-fetch"+ sent.append(request)+ return httpx.Response(200, json=await service.run_hosted_tool(+ "web_fetch", json.loads(request.content), turn=frozen,+ ))++ async with httpx.AsyncClient(transport=httpx.MockTransport(gateway)) as client:+ tools = declared_tools(router(selection_from_hop(hop)), client)+ tool = next(tool for tool in tools if tool.spec.name == "web_fetch")+ output = await tool.execute({"url": "https://example.com/"}, None)+ assert len(sent) == 1 and not output.is_error+ assert isinstance(output.hosted.record, WebFetchRecord)+ message = ToolResultMessage("fetch1", "web_fetch", output.content, hosted=output.hosted)+ saved = message_to_dict(message)+ assert message_from_dict(saved) == message+ assert "local page" in saved["content"][0]["text"]+ assert len(saved["content"][0]["text"].encode("utf-8")) <= 48000+ result = saved["hosted"]assert result["record"]["status"] == "completed"assert result["record"]["attempts"][0]["kind"] == "local"assert result["execution"]["executor_origin"]["api"] == "local_fetch"assert result["execution"]["cost"] == {"basis": "list", "usd": 0}row = next(row for row in service.usage_summary(days=7)["services"] if row["tool"] == "web_fetch")assert row["requests"] == 1 and row["billable_requests"] == 0 and row["uncertain_requests"] == 0+ assert service.store.load().to_payload() == before@pytest.mark.asynciodiff --git a/tests/test_model_hub_hosted_executor.py b/tests/test_model_hub_hosted_executor.pyindex 6583a91ef..f969dbb5b 100644--- a/tests/test_model_hub_hosted_executor.py+++ b/tests/test_model_hub_hosted_executor.py@@ -1121,7 +1121,7 @@ async def test_a_turn_keeps_its_tools_and_executor_order_until_the_next_turn(tmp_bind(service, "web_search", (OPENAI, "gpt-6-astra"))_bind(service, "x_search", (XAI, "grok-4.6"))again = await router.resolve_hop("menu-alias", process_scope="vibey:freeze", turn_id="turn-freeze")- assert again["hosted_tools"].keys() == first["hosted_tools"].keys() == {"web_search", "image_generate"}+ assert again["hosted_tools"].keys() == first["hosted_tools"].keys() == {"web_search", "image_generate", "web_fetch"}status, answer = await search(again, "web_search")assert status == 200 and answer["execution"]["executor_source_id"] == XAIassert (await search(again, "x_search"))[0] == 409diff --git a/tests/test_model_hub_hosted_tools_config.py b/tests/test_model_hub_hosted_tools_config.pyindex 79eaa697c..ceac3ef59 100644--- a/tests/test_model_hub_hosted_tools_config.py+++ b/tests/test_model_hub_hosted_tools_config.py@@ -59,7 +59,7 @@ def test_released_configs_load_with_every_tool_on_and_stay_without_the_section()section = json.loads((FIXTURES / "released_pre_vibey_consumer.json").read_text(encoding="utf-8"))config = ModelHubConfig.from_payload(section, repairing=True)assert config.hosted_tools == {}- assert enabled_tools(config) == RELEASED_TOOLS+ assert enabled_tools(config) == KNOWN_TOOLSassert "hosted_tools" not in config.to_payload()@@ -140,7 +140,7 @@ def test_disk_load_with_a_broken_hosted_tools_section_keeps_model_hub(monkeypatcassert not loaded.load_warningsassert [source.id for source in loaded.model_hub.sources] == ["src_xai00001"]- assert enabled_tools(loaded.model_hub) == ("image_generate",)+ assert enabled_tools(loaded.model_hub) == ("image_generate", "web_fetch")assert config_path.read_text(encoding="utf-8") == originalassert not list(tmp_path.glob("config.json.bak-recovery-*"))@@ -182,7 +182,7 @@ async def test_the_tools_api_writes_one_field_of_one_tool_at_a_time(tmp_path):status = service.hosted_tools_status()["hosted_tools"]_validate_status(status)by_name = {tool["name"]: tool for tool in status["tools"]}- assert list(by_name) == list(RELEASED_TOOLS) and all(tool["enabled"] for tool in status["tools"])+ assert list(by_name) == list(KNOWN_TOOLS) and all(tool["enabled"] for tool in status["tools"])# The executor runs on the vendor's lightweight model, priced by the overlay's dated list price.(candidate,) = by_name["x_search"]["executor_candidates"]assert (candidate["source_id"], candidate["model_id"], candidate["billing"], candidate["relay"]) == (diff --git a/tests/test_model_hub_hosted_tools_resolution.py b/tests/test_model_hub_hosted_tools_resolution.pyindex 2cc47bfc5..27c1a37a6 100644--- a/tests/test_model_hub_hosted_tools_resolution.py+++ b/tests/test_model_hub_hosted_tools_resolution.py@@ -22,6 +22,7 @@ from config.v2_config import (ModelHubModelConfig,ModelHubSourceConfig,ModelHubSourceStateConfig,+ model_hub_builtin_source,)from core.handlers.model_hub import hosted_tools as hosted_tools_modulefrom core.handlers.model_hub.capability_evidence import (@@ -313,13 +314,21 @@ def test_a_turn_freezes_its_tools_executors_and_image_enum_and_each_hop_states_iwith the identity it was built under, and the image enum; a hop states facts only, the enum's params sorted."""evidence = _evidence(tmp_path)turn = freeze_turn_tools(_config(), evidence=evidence, spec_for=_spec_for)- assert turn.tools == {"web_search", "x_search", "image_generate"}+ assert turn.tools == {"web_search", "x_search", "image_generate", "web_fetch"}assert turn.executors == {"web_search": (ExecutorSlot("src_xai00001", "grok-fast", identity_digest(XAI), label="src_xai00001"),ExecutorSlot("src_oai00001", "gpt-mini", identity_digest(OPENAI), label="src_oai00001"),),"x_search": (ExecutorSlot("src_xai00001", "grok-fast", identity_digest(XAI)),),+ "web_fetch": tuple(+ ExecutorSlot(sid, model, identity_digest(model_hub_builtin_source(sid)), kind=kind)+ for sid, model, kind in (+ ("src_localfetch", "local/fetch", "local"),+ ("src_exakeyless", "exa/mcp", "service"),+ ("src_jinakeyless", "jina/reader", "service"),+ )+ ),}image = {"description": "Grok Imagine", "params": ["images", "size"], "size_forms": ["pixels"]}assert hop_hosted_tools(turn, OPENAI, "gpt-6-astra", evidence=evidence, spec_for=_spec_for) == {@@ -329,6 +338,7 @@ def test_a_turn_freezes_its_tools_executors_and_image_enum_and_each_hop_states_i},# OpenAI does not run X search: the hop takes the function."x_search": {"native": "unsupported", "replay": "unsupported"},+ "web_fetch": {"native": "unsupported", "replay": "unsupported"},"image_generate": {"models": [{"source_id": "src_xai00001", "model_id": "grok-imagine-image", **image},@@ -343,7 +353,10 @@ def test_a_turn_freezes_its_tools_executors_and_image_enum_and_each_hop_states_i"native": "unsupported", "replay": "unsupported", "refused": True,}# Off, or with no image model available, a tool is not declared at all.- off = _config({"x_search": ModelHubHostedToolConfig(enabled=False)}, sources=[CLAUDE_SUB, OPENAI, GEMINI])+ off = _config({+ "x_search": ModelHubHostedToolConfig(enabled=False),+ "web_fetch": ModelHubHostedToolConfig(enabled=False),+ }, sources=[CLAUDE_SUB, OPENAI, GEMINI])turn = freeze_turn_tools(off, evidence=_evidence(tmp_path / "o", (OPENAI, "gpt-image-2", "image_generate", "call")), spec_for=_spec_for,)@@ -365,7 +378,7 @@ def test_the_tools_read_states_saved_and_effective_values_and_what_real_use_showconfig = _config({"web_search": ModelHubHostedToolConfig(executor_order=("src_oai00001",))})status = hosted_tools_status(config, evidence=evidence, spec_for=_spec_for, image_price=lambda _source, _spec: None)by_name = {tool["name"]: tool for tool in status["tools"]}- assert list(by_name) == ["web_search", "x_search", "image_generate"]+ assert list(by_name) == ["web_search", "x_search", "image_generate", "web_fetch"]web = by_name["web_search"]assert web["enabled"] is True and by_name["x_search"]["enabled"] is Trueassert web["executor_order"] == ["src_oai00001"]diff --git a/tests/test_model_hub_local_fetch.py b/tests/test_model_hub_local_fetch.pyindex 70866c6b2..6e00d0758 100644--- a/tests/test_model_hub_local_fetch.py+++ b/tests/test_model_hub_local_fetch.py@@ -593,7 +593,8 @@ async def test_stop_during_success_cleanup_is_rethrown_after_reap(monkeypatch):await task-async def test_active_worker_cap_and_cancellation_safe_host_release(monkeypatch):[email protected]("spawn_delay", [0, 0.4], ids=["normal-start", "slow-start"])+async def test_active_worker_cap_and_cancellation_safe_host_release(monkeypatch, spawn_delay):script = '''import sys,time,jsonfrom urllib.parse import urlsplitrequest=json.loads(sys.stdin.readline())@@ -602,9 +603,12 @@ print('{"event":"sent"}',flush=True); sys.stdin.readline()time.sleep(60)'''monkeypatch.setattr(local_fetch, "_WORKER", script)- children, spawn = [], asyncio.create_subprocess_exec+ children, spawn, spawn_attempts = [], asyncio.create_subprocess_exec, []reaping, release = asyncio.Event(), asyncio.Event()async def capture(*args, **kwargs):+ spawn_attempts.append(1)+ if spawn_delay:+ await asyncio.sleep(spawn_delay)child = await spawn(*args, **kwargs)children.append(child)if len(children) == 1:@@ -634,11 +638,21 @@ time.sleep(60)again.cancel()await asyncio.gather(again, return_exceptions=True)children.clear()+ spawn_attempts.clear()# Distinct hosts still have one aggregate process cap.- tasks = [asyncio.create_task(engine.fetch(f"https://host{i}.example/", deadline=time.monotonic() + 10)) for i in range(20)]+ sent, ready = 0, asyncio.Event()+ worker_cap = local_fetch._WORKERS+ def mark_sent():+ nonlocal sent+ sent += 1+ if sent == worker_cap:+ ready.set()+ tasks = [asyncio.create_task(engine.fetch(f"https://host{i}.example/", deadline=time.monotonic() + 10,+ on_sent=mark_sent)) for i in range(20)]try:- await asyncio.sleep(.3)- assert 1 <= len(children) <= 8+ await asyncio.wait_for(ready.wait(), 10)+ assert sent == worker_cap+ assert len(spawn_attempts) == len(children) == worker_cap == 8finally:for task in tasks:task.cancel()diff --git a/tests/test_model_hub_vibey_consumer.py b/tests/test_model_hub_vibey_consumer.pyindex 83d4bc9d6..27007d525 100644--- a/tests/test_model_hub_vibey_consumer.py+++ b/tests/test_model_hub_vibey_consumer.py@@ -1167,6 +1167,8 @@ async def test_hop_resolution_states_each_enabled_hosted_tool(tmp_path):assert resolving_threads and threading.main_thread() not in resolving_threads_validate("hop-resolution.schema.json", hop)assert hop["hosted_tools"] == {+ # Fetch is enabled by default, but is function-only on every model.+ "web_fetch": {"native": "unsupported", "replay": "unsupported"},"web_search": {"native": "assumed","declaration": {"type": "web_search"},diff --git a/tests/test_model_hub_web_fetch_readers.py b/tests/test_model_hub_web_fetch_readers.pyindex a94610820..8aba269fe 100644--- a/tests/test_model_hub_web_fetch_readers.py+++ b/tests/test_model_hub_web_fetch_readers.py@@ -1,4 +1,4 @@-"""WF-1: dormant settings, independent order, and future ledger readers.+"""WF-1/WF-3: saved fetch consent, active reads and independent ledger attribution.The existing search cases cannot catch erased fetch consent or merged Exa toolrows. These tests exercise the settings/read and ledger owners, not a producer.@@ -38,7 +38,7 @@ def fetch_config(settings=None):@pytest.mark.asyncioasync def test_fetch_consent_survives_v2_reload_and_unrelated_save(tmp_path):- """WF-1: a dormant build must not erase a newer build's built-in opt-outs."""+ """Fetch activation must not erase persisted built-in opt-outs on unrelated saves."""config = fetch_config()assert config.to_payload()["hosted_tools"]["web_fetch"] == FETCH_SETTINGSservice, path, _ = _disk_service(tmp_path, config)@@ -94,8 +94,8 @@ def test_future_fetch_ledger_rows_do_not_merge_shared_exa_identity(tmp_path, winassert {(r.get("tool", "web_search"), r["requests"]) for r in reopened["services"]} == {("web_search", 3), ("web_fetch", 3)}-def test_dormant_hop_reader_accepts_function_only_fetch_but_never_declares_it():- """WF-1 can read a future C-6 entry without enabling an execution path."""+def test_hop_reader_declares_function_only_fetch_by_default():+ """WF-3: a supplied C-6 entry registers the canonical function without flag patches."""from types import SimpleNamespacefrom modules.agents.model_hub import ModelHubLaunchfrom modules.agents.vibey.models import selection_from_hop@@ -106,17 +106,21 @@ def test_dormant_hop_reader_accepts_function_only_fetch_but_never_declares_it():hop["hosted_tools"] = {"web_fetch": {"native": "unsupported", "replay": "unsupported"}}selection = selection_from_hop(hop)assert selection.endpoint.hosted_tools["web_fetch"].mode == "proxy"- assert turn_tools(SimpleNamespace(preflight=selection, current=selection)) == ()+ assert [tool.spec.name for tool in turn_tools(SimpleNamespace(preflight=selection, current=selection))] == ["web_fetch"]-def test_dormant_read_candidates_and_turn_do_not_advertise_fetch(tmp_path):- """A known stored tool must not implicitly enter a live turn or the Tools UI."""- service, path, _ = _disk_service(tmp_path, fetch_config())+def test_active_read_candidates_and_turn_preserve_saved_fetch_consent(tmp_path):+ """The visible row and frozen candidates agree without rewriting saved choices."""+ config = fetch_config()+ service, path, _ = _disk_service(tmp_path, config)before = path.read_bytes()- assert hosted_tools.WEB_FETCH_ACTIVE is False- assert "web_fetch" not in {t["name"] for t in service.hosted_tools_status()["hosted_tools"]["tools"]}- assert "web_fetch" not in hosted_tools.freeze_turn_tools(fetch_config(), evidence=service.hosted_evidence(), spec_for=hosted_tools.hosted_tool_spec).tools- assert not hosted_tools.executor_candidates(fetch_config(), "web_fetch", evidence=service.hosted_evidence(), spec_for=hosted_tools.hosted_tool_spec)+ fetch = next(t for t in service.hosted_tools_status()["hosted_tools"]["tools"] if t["name"] == "web_fetch")+ turn = service.freeze_turn_hosted_tools(config)+ assert "web_fetch" in turn.tools+ assert [slot.source_id for slot in turn.executors["web_fetch"]] == fetch["effective_order"]+ assert {slot.source_id: slot.enabled for slot in turn.executors["web_fetch"]} == {+ LOCAL: False, FREE_EXA: False, JINA: False, EXA: True, TAVILY: True, DEEPSEEK: True,+ }assert path.read_bytes() == before@@ -127,10 +131,9 @@ def test_dormant_read_candidates_and_turn_do_not_advertise_fetch(tmp_path):({"executor_order": [LOCAL, JINA], "executor_order_kinds": {LOCAL: "service", JINA: "service"}}, [LOCAL, EXA, TAVILY, DEEPSEEK, JINA, FREE_EXA]),({"services": {LOCAL: {"enabled": False}, EXA: {"enabled": False}}}, [TAVILY, DEEPSEEK, FREE_EXA, JINA, LOCAL, EXA]),])-def test_active_fetch_read_owns_independent_order_without_writing(tmp_path, monkeypatch, settings, expected):+def test_active_fetch_read_owns_independent_order_without_writing(tmp_path, settings, expected):"""Search's disabled Tavily/order cannot decide fetch consent or deliberate placement."""from tests.test_model_hub_api import _api_response_registry- monkeypatch.setattr(hosted_tools, "WEB_FETCH_ACTIVE", True)service, path, _ = _disk_service(tmp_path, fetch_config(settings))before = path.read_bytes()payload = service.hosted_tools_status()["hosted_tools"]@@ -164,12 +167,11 @@ def test_fetch_catalog_capability_and_catalog_off_preserve_identity(monkeypatch)@pytest.mark.asyncio@pytest.mark.parametrize("sid,service_id", [(FREE_EXA, "exa/mcp"), (JINA, "jina/reader")])-async def test_synthetic_fetch_cooldown_survives_save_and_retry_is_shared(tmp_path, monkeypatch, sid, service_id):+async def test_synthetic_fetch_cooldown_survives_save_and_retry_is_shared(tmp_path, sid, service_id):"""Retry uses the same synthetic identity/generation as other tools; no request is made."""from config.v2_config import model_hub_builtin_sourcefrom core.handlers.model_hub.search_services import ServiceCooldowns, ServiceResultfrom core.handlers.model_hub.adapter import SearchResponse- monkeypatch.setattr(hosted_tools, "WEB_FETCH_ACTIVE", True)service, _, _ = _disk_service(tmp_path, fetch_config({}))service.now = lambda: NOWservice._service_cooldowns = ServiceCooldowns()@@ -191,7 +193,7 @@ async def test_synthetic_fetch_cooldown_survives_save_and_retry_is_shared(tmp_pa@pytest.mark.asyncio@pytest.mark.parametrize("sid,service_id,label", [(LOCAL, "local/fetch", "Local"), (FREE_EXA, "exa/mcp", "Exa (free)"), (JINA, "jina/reader", "Jina Reader (free)")])async def test_synthetic_fetch_evidence_and_usage_survive_without_credential_sources(tmp_path, sid, service_id, label):- """Synthetic identities join both durable readers even when activation is off."""+ """Synthetic identities join both durable readers without becoming credential Sources."""from config.v2_config import model_hub_builtin_sourcefrom core.handlers.model_hub.capability_evidence import CapabilityEvidenceStoreservice, _, _ = _disk_service(tmp_path, fetch_config())diff --git a/tests/test_vibey_agent.py b/tests/test_vibey_agent.pyindex c08b8ebbf..c2d81350a 100644--- a/tests/test_vibey_agent.py+++ b/tests/test_vibey_agent.py@@ -3915,6 +3915,131 @@ async def test_a_proxied_hosted_call_runs_on_the_turns_executor_chain_and_keeps_assert sent["type"] == "tool_result" and seen in json.dumps(sent)[email protected]("platform", ["avibe", "telegram"])[email protected]("executor", ["local", "exa", "jina", "policy"])+async def test_web_fetch_survives_the_real_agent_commit_delivery_and_next_turn_replay(+ engine, session, tmp_path, published, monkeypatch, platform, executor+):+ """WF-3: a gateway 200 must commit and display, not become an interrupted call.++ The earlier tool-only acceptance never reached SQLite media registration.+ Exercise the real parser, Agent loop, tool runtime, transcript, dispatcher,+ mirror and activity reader; only HTTP is stubbed. A fresh adapter then replays+ the stored fetch with the tool disabled, without another fetch or media write.+ """+ import httpx++ from core.agent_core.messages import ServiceExecutorOrigin, WebFetchRecord+ from core.agent_core.web_fetch import frame_fetch_record+ from modules.agents.vibey.models import registry_providers+ from storage import agent_activity_service+ from storage.models import media_objects++ monkeypatch.setattr("core.message_mirror._activity_streaming_enabled", lambda: True)+ blocked = executor == "policy"+ url = "http://10.0.0.5/" if blocked else "https://example.test/page"+ content = "" if blocked else "Hello 世界"+ sid, label, kind, vendor, service_id = {+ "local": ("src_localfetch", "Local", "local", "local", "local/fetch"),+ "exa": ("src_exakeyless", "Exa (free)", "service", "exa-mcp", "exa/mcp"),+ "jina": ("src_jinakeyless", "Jina Reader (free)", "service", "jina-reader", "jina/reader"),+ "policy": (None, None, None, None, None),+ }[executor]+ cost = HostedCost(basis="list", usd=0)+ record = WebFetchRecord(+ status="failed" if blocked else "completed", requested_url=url,+ content=content, size_bytes=len(content.encode("utf-8")), retrieved_at="2026-10-25T17:27:45Z",+ error="blocked by local policy" if blocked else None,+ error_code="fetch_url_blocked" if blocked else None,+ attempts=() if blocked else (HostedAttempt(sid, label, kind, "completed", cost=cost),),+ )+ hosted = HostedResult(record=record, execution=None if blocked else ProxyExecution(+ executor_origin=ServiceExecutorOrigin(vendor, "local_fetch" if kind == "local" else "fetch_api", service_id),+ executor_source_id=sid, executor_billing="metered", cost=cost,+ ))+ frame = frame_fetch_record(record)+ expected = ToolResultMessage("toolu_fetch", "web_fetch", (text(frame),), is_error=blocked, hosted=hosted)+ envelope = message_to_dict(expected)["hosted"]+ served = {"provider": "test-provider", "api": "anthropic", "model": "test-model", "source": "src_served00001"}+ replies = [+ _anthropic_reply(json.dumps({"type": "tool_use", "id": "toolu_fetch", "name": "web_fetch", "input": {"url": url}}), stop_reason="tool_use"),+ _anthropic_reply('{"type":"text","text":"fetch handled"}', stop_reason="end_turn"),+ _anthropic_reply('{"type":"text","text":"stored read recalled"}', stop_reason="end_turn"),+ ]+ models, executed = [], []++ def gateway(request):+ if request.url.path == "/vibey/v1/hosted-tools/web_fetch":+ executed.append(request)+ assert json.loads(request.content) == {"arguments": {"url": url}}+ return httpx.Response(200, json=envelope)+ assert request.url.path == "/vibey/v1/messages"+ models.append(json.loads(request.content))+ return httpx.Response(200, headers={+ "content-type": "text/event-stream", "x-avibe-served-hop": json.dumps(served),+ }, text=replies.pop(0))++ async with httpx.AsyncClient(transport=httpx.MockTransport(gateway)) as client:+ harness = _Harness(engine, tmp_path, platform, [],+ providers=registry_providers(media_loader=None, client=client), gateway_client=client)+ harness.controller.hub_source_id = served["source"]+ harness.controller.hub_hosted_tools = {"web_fetch": {"native": "unsupported", "replay": "unsupported"}}+ harness.controller.im_client.supports_message_editing = lambda context=None: False+ first = harness.request("read this page")+ await harness.agent.handle_message(first)+ assert harness.controller.terminals[-1]["is_error"] is False+ assert harness.controller.terminal_texts[-1] == "fetch handled"+ [committed] = [row for row in await harness.context_rows() if row.kind == "tool_result"]+ assert committed.message == expected+ assert executed[0].headers["x-avibe-turn-id"] == _turn(first.context)+ assert executed[0].headers["authorization"] == "Bearer hub-token"++ # Reconstruct from durable rows, with no web_fetch declaration on the new turn.+ harness.new_agent()+ harness.controller.hub_hosted_tools = {}+ await harness.agent.handle_message(harness.request("recall the stored read"))+ assert harness.controller.terminals[-1]["is_error"] is False+ assert harness.controller.terminal_texts[-1] == "stored read recalled"+ [replayed] = [row for row in await harness.context_rows() if row.kind == "tool_result"]+ assert replayed == committed++ assert len(executed) == 1 and len(models) == 3 and replies == []+ assert models[0]["tools"][0]["name"] == "web_fetch"+ assert all(tool.get("name") != "web_fetch" for tool in models[2].get("tools", []))+ for request in models[1:]:+ [answer] = [block for message in request["messages"] for block in message["content"]+ if block.get("type") == "tool_result" and block.get("tool_use_id") == "toolu_fetch"]+ assert answer["content"] == [{"type": "text", "text": frame}]+ assert answer.get("is_error", False) is blocked+ assert "tool call interrupted" not in json.dumps(request)++ # The dispatcher/mirror create exactly one hosted trace, read unchanged by+ # the activity API. Page records never register images or produce uploads.+ with engine.connect() as conn:+ traces = conn.execute(select(agent_events).where(+ agent_events.c.session_id == SESSION, agent_events.c.event_type == "tool_call"+ )).mappings().all()+ groups = agent_activity_service.list_turn_groups(conn, session_id=SESSION)["groups"]+ details = [agent_activity_service.get_turn_group(conn, session_id=SESSION, group_id=group["id"]) for group in groups]+ assert conn.execute(select(media_objects.c.token).where(media_objects.c.session_id == SESSION)).first() is None+ [trace] = [row for row in traces if json.loads(row["content_json"]).get("hosted")]+ card = json.loads(trace["content_json"])["hosted"]+ assert card["tool"] == "web_fetch" and card["status"] == record.status+ assert card["sources"] == card["images"] == []+ assert card["fetch"] == envelope["record"]+ if blocked:+ assert card["policy_sources"] == [] and card["error_code"] == "fetch_url_blocked"+ else:+ assert card["source_id"] == sid and card["cost"]["usd"] == 0+ assert [row["hosted"] for detail in details for row in detail["rows"] if row.get("hosted")] == [card]+ live = [payload["content"]["hosted"] for name, payload in published+ if name == "message.new" and (payload.get("content") or {}).get("hosted")]+ assert live == ([card] if platform == "avibe" else [])+ assert harness.controller.im_client.sent.count(trace["content_text"]) == 1+ assert not harness.controller.im_client.uploads+ assert len(harness.rows("result")) == 2 and not harness.rows("error")++_JPEG = b"\xff\xd8\xff\xe0" + bytes(range(256)) * 8_DIRECT = {"mode": "direct",diff --git a/tests/test_vibey_web_fetch.py b/tests/test_vibey_web_fetch.pyindex ab744212a..d3c18b433 100644--- a/tests/test_vibey_web_fetch.py+++ b/tests/test_vibey_web_fetch.py@@ -46,7 +46,10 @@ def router(first):@pytest.mark.parametrize("protocol", ["anthropic", "openai_responses", "openai_chat", "google"])@pytest.mark.parametrize("active,offered", [(False, True), (True, False), (True, True)])async def test_fetch_declaration_is_a_frozen_url_only_function(monkeypatch, protocol, active, offered):- monkeypatch.setattr(v2_config, "WEB_FETCH_ACTIVE", active)+ if not active:+ monkeypatch.setattr(v2_config, "WEB_FETCH_ACTIVE", False)+ else:+ assert v2_config.WEB_FETCH_ACTIVE is Trueturn = router(selection(protocol, offered=offered))tools = hosted_tools(turn)assert [tool.spec.name for tool in tools] == (["web_fetch"] if active and offered else [])@@ -72,8 +75,7 @@ async def test_fetch_declaration_is_a_frozen_url_only_function(monkeypatch, prot@pytest.mark.parametrize("outcome", ["completed", "policy", "closed", "off", "malformed", "unreachable"])-async def test_gateway_result_preserves_one_frame_and_live_turn(monkeypatch, outcome):- monkeypatch.setattr(v2_config, "WEB_FETCH_ACTIVE", True)+async def test_gateway_result_preserves_one_frame_and_live_turn(outcome):stored = json.loads(FIXTURE.read_text())answer = copy.deepcopy(stored["hosted"])if outcome == "policy":@@ -119,8 +121,7 @@ async def test_gateway_result_preserves_one_frame_and_live_turn(monkeypatch, out@pytest.mark.parametrize("length", [8192, 8193])-async def test_gateway_failure_never_fabricates_a_shortened_requested_url(monkeypatch, length):- monkeypatch.setattr(v2_config, "WEB_FETCH_ACTIVE", True)+async def test_gateway_failure_never_fabricates_a_shortened_requested_url(length):url = "https://example.test/" + "a" * (length - len("https://example.test/"))def unavailable(request):raise httpx.ConnectError("offline", request=request)@@ -134,8 +135,7 @@ async def test_gateway_failure_never_fabricates_a_shortened_requested_url(monkey@pytest.mark.parametrize("protocol", ["anthropic", "openai_responses", "openai_chat", "google"])-async def test_newly_produced_result_replays_without_fetch_or_rebill(monkeypatch, protocol):- monkeypatch.setattr(v2_config, "WEB_FETCH_ACTIVE", True)+async def test_newly_produced_result_replays_without_fetch_or_rebill(protocol):fixture = json.loads(FIXTURE.read_text())calls = []def gateway(request):