feat(harness): add an AI SDK harness by cjol · Pull Request #2434 · cloudflare/agents

cjol · 2026-10-01T13:55:52Z

This PR adds agents/harness/ai-sdk: AiSdkHarness, which serves an AI SDK model through Channels, and the AI SDK conversions it is built on.

Why

Channels serves a harness. An agent written against the AI SDK has no harness of its own, so without one every agent would rebuild sessions, a transcript, a turn queue and recovery around streamText.
An AI SDK agent speaks UIMessage and UIMessageChunk. Channels speaks TranscriptMessage and ResponseChunk. The conversions, including tool states and approvals, live with the harness that needs them.
The harness is opt-in and knows nothing about Channels setup. It implements the shared interface, as the pi adapter does, and sits next to agents/harness/pi.

Public API Surface

Additive, from agents/harness/ai-sdk:

Symbol	Kind	Notes
`AiSdkHarness`, `AiSdkHarnessOptions`	class, type	Lifecycle capability: `model`, `tools`, `system`, `stopWhen`, `heartbeatMs`
`toResponseChunks`	function	Converts an AI SDK stream to response chunks
`toTranscriptMessage`, `toUIMessage`	functions	Message conversions in each direction
`awaitsInput`, `answerToolCall`, `ToolCallAnswer`	functions, type	Approvals and client tool results
`createSendMessageTool`, `CreateSendMessageToolOptions`	function, type	AI SDK tool that sends to a surface through the gateway
`AiSdkConversionOptions`	type	Conversion options

Code Changes

harness/ai-sdk/harness.ts keeps sessions, transcripts and an operation queue in the object's storage. Each operation runs one streamText call over the session's transcript, and a Lifecycle wake job runs the queue again after an eviction. A client tool takes its result only from the participant who started the turn, while approvals come from anyone. A steer queues like a follow-up.
harness/ai-sdk/turns.ts holds the conversions, and send-message-tool.ts the send tool.
New entry point in package.json and scripts/build.ts, and a minor changeset for agents.
ai-sdk-harness.test.ts runs the harness behind Channels and the Web Channel with a mock model.

changeset-bot · 2026-10-01T13:56:04Z

🦋 Changeset detected

Latest commit: fec4998

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 2 packages

Name	Type
agents	Minor
@cloudflare/agent-think	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

agent-think · 2026-10-01T13:59:23Z

🟢 agents import sizes: 1 entry point changed, no growth

	Entry point	Exports	Largest gzip change	Size now
🆕	`agents/harness/ai-sdk`	7 new	—	157.4 KiB

Changed exports (7)

	Import	Gzip change	Size now
🆕	`agents/harness/ai-sdk#AiSdkHarness`	—	157.4 KiB
🆕	`agents/harness/ai-sdk#toResponseChunks`	—	112.8 KiB
🆕	`agents/harness/ai-sdk#toTranscriptMessage`	—	112.7 KiB
🆕	`agents/harness/ai-sdk#answerToolCall`	—	112.4 KiB
🆕	`agents/harness/ai-sdk#toUIMessage`	—	112.3 KiB
🆕	`agents/harness/ai-sdk#awaitsInput`	—	112.2 KiB
🆕	`agents/harness/ai-sdk#createSendMessageTool`	—	112.2 KiB

How this works

Each runtime export is bundled on its own, minified, and gzipped. Changes smaller than 100 B, or smaller than 1% and 1 KiB, are ignored. Growth over 10% or 5 KiB is marked 🔴. This report is informational and does not fail CI. The workflow artifact contains every measurement.

_{Compared 1b17acd1 → fec4998a · workflow run · reported by agent-think[bot]}

pkg-pr-new · 2026-10-01T14:42:48Z

Open in StackBlitz

agents

npm i https://pkg.pr.new/agents@2434

@cloudflare/ai-chat

npm i https://pkg.pr.new/@cloudflare/ai-chat@2434

@cloudflare/codemode

npm i https://pkg.pr.new/@cloudflare/codemode@2434

hono-agents

npm i https://pkg.pr.new/hono-agents@2434

@cloudflare/shell

npm i https://pkg.pr.new/@cloudflare/shell@2434

@cloudflare/think

npm i https://pkg.pr.new/@cloudflare/think@2434

@cloudflare/voice

npm i https://pkg.pr.new/@cloudflare/voice@2434

@cloudflare/worker-bundler

npm i https://pkg.pr.new/@cloudflare/worker-bundler@2434

commit: fec4998

devin-ai-integration

Devin Review found 4 new potential issues.

devin-ai-integration · 2026-10-02T17:39:56Z

#reset

+    for (const [key] of this.#kv.list({ prefix: messagePrefix(session) })) {
+      this.#kv.delete(key);
+    }


🔴 Reset blocks replay of old messages

After reset, resending an earlier message fails because #operation still finds its settled operation ID. Channels clears its event cache for replay, but the transcript cannot be rebuilt.

Learn more

Reset removes transcript entries but leaves operation records and their deduplication keys in durable storage. Channels explicitly clears its own seen-event cache on reset so a previously accepted event can be resent to rebuild the conversation in #reset. When that event uses its original ID, #submit finds the old operation and rejects it without placing a new message.

Example: A message with event ID e1 completes. The user resets the conversation and resends e1; Channels accepts it again, but the harness returns accepted: false, leaving the new transcript empty.

Recommended fix: Clear or scope settled operation records and sequence/index state to the reset generation, while retaining any deduplication required for operations still being aborted. Verify that replaying an old event ID after reset places its message and starts a new run.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-10-02T17:39:57Z

answerToolCall

+      const holder = messages.find((m) =>
+        m.parts.some(
+          (p) => "toolCallId" in p && p.toolCallId === answer.toolCallId
+        )
+      );
+      const owner = holder?.metadata?.owner;
+      const from = answer.from?.participantId;
+      if (owner !== undefined && from !== undefined && owner !== from) {


🟡 Reused tool IDs reject valid answers

When a later client tool reuses a tool ID, #answer checks the first historical call's owner. A different participant's valid result gets not_owner, leaving the current call unanswered.

Learn more

Tool call IDs come from the model and are not namespaced by session turn. The ownership lookup searches all saved messages in order, without checking whether the call still awaits a result. By contrast, answerToolCall selects a call by both ID and eligible state. When an old completed call and a current pending call share an ID, their owners can differ, and this ownership check rejects the current owner's answer.

Example: Alice answers tool call c1 in the first turn. Bob starts a later turn whose model also emits c1. Bob's answer sees Alice's completed call first and is refused as not_owner.

Recommended fix: Locate the same pending tool part that answerToolCall will update, then check that message's owner. Prefer one shared matching helper so the ownership check and update cannot select different calls.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-10-02T17:39:58Z

wait contract

+          waiters.add(done);
+          signal?.addEventListener("abort", aborted, { once: true });


🟡 Pre-aborted waits remain pending

When wait receives an already aborted signal, it registers a listener after the abort event. The wait cannot cancel until the operation settles, leaving callers blocked.

Learn more

An AbortSignal dispatches its abort event only once. If wait(operationId, signal) starts with a pending operation and an already aborted signal, the newly added listener never runs; only operation settlement resolves the promise. The shared wait contract accepts a signal for cancellation.

Example: An operation waits on a slow model and a caller passes AbortSignal.abort() to wait. Instead of rejecting immediately, its promise stays pending until the model run ends.

Recommended fix: Check signal.aborted before registering the waiter and reject with signal.reason. Also remove the waiter and its map entry when an active signal aborts.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-10-02T17:39:59Z

+    "./harness/ai-sdk": {
+      "types": "./dist/harness/ai-sdk/index.d.ts",
+      "import": "./dist/harness/ai-sdk/index.js",
+      "require": "./dist/harness/ai-sdk/index.js"
+    },


🔍 Advertised import path differs from exported API

The PR description advertises agents/channels/ai-sdk, but the package exports agents/harness/ai-sdk and a hosted harness. Align the public API documentation with the intended release.

Was this helpful? React with 👍 or 👎 to provide feedback.

This PR adds `agents/harness/ai-sdk`: `AiSdkHarness`, which serves an AI SDK model through Channels, and the AI SDK conversions it is built on. ## Why - Channels serves a harness. An agent written against the AI SDK has no harness of its own, so without one every agent would rebuild sessions, a transcript, a turn queue and recovery around `streamText`. - An AI SDK agent speaks `UIMessage` and `UIMessageChunk`. Channels speaks `TranscriptMessage` and `ResponseChunk`. The conversions, including tool states and approvals, live with the harness that needs them. - The harness is opt-in and knows nothing about Channels setup. It implements the shared interface, as the pi adapter does, and sits next to `agents/harness/pi`. ## Public API Surface Additive, from `agents/harness/ai-sdk`: | Symbol | Kind | Notes | | --- | --- | --- | | `AiSdkHarness`, `AiSdkHarnessOptions` | class, type | Lifecycle capability: `model`, `tools`, `system`, `stopWhen`, `heartbeatMs` | | `toResponseChunks` | function | Converts an AI SDK stream to response chunks | | `toTranscriptMessage`, `toUIMessage` | functions | Message conversions in each direction | | `awaitsInput`, `answerToolCall`, `ToolCallAnswer` | functions, type | Approvals and client tool results | | `createSendMessageTool`, `CreateSendMessageToolOptions` | function, type | AI SDK tool that sends to a surface through the gateway | | `AiSdkConversionOptions` | type | Conversion options | ## Code Changes - `harness/ai-sdk/harness.ts` keeps sessions, transcripts and an operation queue in the object's storage. Each operation runs one `streamText` call over the session's transcript, and a Lifecycle wake job runs the queue again after an eviction. A client tool takes its result only from the participant who started the turn, while approvals come from anyone. A steer queues like a follow-up. - `harness/ai-sdk/turns.ts` holds the conversions, and `send-message-tool.ts` the send tool. - New entry point in `package.json` and `scripts/build.ts`, and a minor changeset for `agents`. - `ai-sdk-harness.test.ts` runs the harness behind Channels and the Web Channel with a mock model.

## pnpm-workspace.yaml (default) ## Dependency Updates | Package | From | To | Type | | --- | --- | --- | --- | | `agents` | 0.26.0 | 0.27.0 | minor | ## Release Notes <details> <summary><b>agents</b> (0.26.0 → 0.27.0)</summary> This release adds four new experimental harnesses and a `web_search` tool, and reorganizes the experimental Channels API. - **New experimental harnesses:** `AiSdkHarness`, `ThinkHarness`, `ContainerHarness` (which runs Claude Code or Codex in a Container) and `OpenCodeHarness`, all with the same shape as `PiHarness`. - **New `agents/websearch`:** a `web_search` tool over Cloudflare's Web Search API for pi, the AI SDK and TanStack AI. - **Channels moved to `agents/experimental/channels`:** Channels is being rebuilt around conversations and turns, and the old `agents/channels` entry point has been removed. The new API includes `ChannelGateway`, the Web Channel and its client, an AI SDK chat transport, and a new `npx agents tui` terminal client. - **Fixes:** Streams, Tasks, `useAgentChat` reconnects, `WebChannelClient`, MCP credential cleanup on `removeServer()`, and x402 with MCP SDK v2. ### Minor Changes - [#2434](cloudflare/agents#2434) [`0ebeae5`](cloudflare/agents@0ebeae5) Thanks [@cjol](https://github.com/cjol)! - Add `AiSdkHarness` from the new `agents/harness/ai-sdk` entry point, which runs each message with `streamText` and keeps sessions and transcripts in the Durable Object, so `Channels.forHarness` can serve an AI SDK model. The same entry point exports the AI SDK message conversions it is built on and `createSendMessageTool`, an AI SDK tool that sends a message to a surface through the gateway. - [#2431](cloudflare/agents#2431) [`6393265`](cloudflare/agents@6393265) Thanks [@cjol](https://github.com/cjol)! - Add `Channels` to `agents/experimental/channels`: a Lifecycle capability that serves an agent harness's sessions as conversations to every surface that joins them, with durable, resumable responses on `Streams`. Connect a harness with `Channels.forHarness(harness, { channel …[full notes](https://github.com/cloudflare/agents/releases/tag/agents%400.27.0) </details> --- *This PR was auto-generated by [catalog-update-action](https://github.com/brandhaug/catalog-update-action).* Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

cjol added this pull request to stack #2440 October 1, 2026 13:56

This comment was marked as resolved.

Sign in to view

cjol force-pushed the channels/06-ai-sdk-connectors branch 2 times, most recently from c1b3768 to e308e8c Compare October 1, 2026 14:27

cjol force-pushed the channels/06-ai-sdk-connectors branch from e308e8c to af3f181 Compare October 2, 2026 10:13

cjol force-pushed the channels/06-ai-sdk-connectors branch from af3f181 to 6ab1014 Compare October 2, 2026 11:08

cjol removed this pull request from stack #2440 October 2, 2026 17:37

cjol force-pushed the channels/06-ai-sdk-connectors branch from 6ab1014 to 245a7b9 Compare October 2, 2026 17:37

cjol changed the title ~~feat(channels): add AI SDK agent connectors~~ feat(harness): add an AI SDK harness Oct 2, 2026

cjol added this pull request to stack #2460 October 2, 2026 17:39

devin-ai-integration Bot reviewed Oct 2, 2026

View reviewed changes

cjol force-pushed the channels/06-ai-sdk-connectors branch from 245a7b9 to 22f8aa5 Compare October 2, 2026 18:20

cjol force-pushed the channels/06-ai-sdk-connectors branch from 22f8aa5 to fec4998 Compare October 2, 2026 18:53

cjol merged commit 0ebeae5 into main Oct 2, 2026
16 of 30 checks passed

cjol deleted the channels/06-ai-sdk-connectors branch October 2, 2026 19:20

github-actions Bot mentioned this pull request Oct 2, 2026

Version Packages #2461

Merged

cjol mentioned this pull request Oct 5, 2026

fix(harness): AiSdkHarness wait rejects for an already aborted signal #2480

Merged

brandhaug mentioned this pull request Oct 10, 2026

chore(deps): bump agents from 0.26.0 to 0.27.0 brandhaug/b2b-saas-starter#669

Merged

Navigation Menu

Repository navigation

feat(harness): add an AI SDK harness - #2434

feat(harness): add an AI SDK harness#2434
cjol merged 1 commit into
channels/05-webfrom
channels/06-ai-sdk-connectors

cjol commented Oct 1, 2026 •

edited

Loading

Uh oh!

changeset-bot Bot commented Oct 1, 2026 •

edited

Loading

Uh oh!

This comment was marked as resolved.

Uh oh!

agent-think Bot commented Oct 1, 2026 •

edited

Loading

Uh oh!

pkg-pr-new Bot commented Oct 1, 2026 •

edited

Loading

Uh oh!

devin-ai-integration Bot left a comment

Uh oh!

devin-ai-integration Bot Oct 2, 2026

Uh oh!

devin-ai-integration Bot Oct 2, 2026

Uh oh!

devin-ai-integration Bot Oct 2, 2026

Uh oh!

devin-ai-integration Bot Oct 2, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		waiters.add(done);
		signal?.addEventListener("abort", aborted, { once: true });

Conversation

cjol commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why

Public API Surface

Code Changes

Uh oh!

changeset-bot Bot commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

This comment was marked as resolved.

Uh oh!

agent-think Bot commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🟢 agents import sizes: 1 entry point changed, no growth

Uh oh!

pkg-pr-new Bot commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

devin-ai-integration Bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Oct 2, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Oct 2, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Oct 2, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Oct 2, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

cjol commented Oct 1, 2026 •

edited

Loading

changeset-bot Bot commented Oct 1, 2026 •

edited

Loading

agent-think Bot commented Oct 1, 2026 •

edited

Loading

pkg-pr-new Bot commented Oct 1, 2026 •

edited

Loading