Repository navigation

Issues

New issue

Project Think roadmap: durable agent platform preview to stable developer experience
#1439 · threepointone opened on May 1, 2026
1

Agent readiness scan aborts homepage GET despite successful HTTP 200 responses
cloudflare/agents#2882 · opened on Nov 13, 2026
1
steering-message chunks from AI SDK steering are dropped by applyChunkToParts and never persisted
cloudflare/agents#2878 · opened on Nov 12, 2026
1
Compaction overlay stops applying once the thread passes ~10k rows
cloudflare/agents#2871 · opened on Nov 12, 2026
agents/streams: list() newest-first tiebreak, async cutover commit, unvalidated read inputs
cloudflare/agents#2853 · opened on Nov 10, 2026
1
Think approval turn waits ~30 s (deadman alarm) to recover when the DO is reset during the alarm job
cloudflare/agents#2847 · opened on Nov 8, 2026
3
Several models used in examples/docs are being deprecated on Nov 20
cloudflare/agents#2844 · opened on Nov 7, 2026
1
@cloudflare/think 0.23.0 breaks Workers builds on just-bash, which nothing installs.
cloudflare/agents#2842 · opened on Nov 6, 2026
2
Lifecycle.fetch returns the error's stack to the caller when startup or onRequest throws
cloudflare/agents#2828 · opened on Nov 5, 2026
2
Streamable HTTP initialize makes an unused getInitializeRequest round trip
cloudflare/agents#2813 · opened on Nov 4, 2026
1
worker-bundler: subpath imports to a directory with its own package.json resolve as external
cloudflare/agents#2810 · opened on Nov 3, 2026
createAI: string catalog models drop Anthropic adaptive thinking and effort providerOptions
cloudflare/agents#2809 · opened on Nov 3, 2026
1
ThinkChat sends the full transcript on every message update, and get-messages has no paging
cloudflare/agents#2788 · opened on Nov 2, 2026
1
OpenCodeHarness: wait() and prompt() pause 1 s after every run
cloudflare/agents#2784 · opened on Oct 31, 2026
1
Think: an interrupted continuation recovered twice is persisted as two assistant rows with the same toolCallId
cloudflare/agents#2780 · opened on Oct 30, 2026
1
agents/models/opencode: OpenCode needs to let plugins supply an AI SDK provider on workerd
cloudflare/agents#2777 · opened on Oct 29, 2026
1
agents/context: cache the system prompt by default and restore the fluent withContext API
cloudflare/agents#2772 · opened on Oct 29, 2026
2
Detached runs: a given-up run whose child was torn down stays outstanding forever, so _cfDetachedReconcileTick re-arms indefinitely
cloudflare/agents#2768 · opened on Oct 29, 2026
1
createAI: support turning reasoning off on Workers AI-backed dynamic routes
cloudflare/agents#2754 · opened on Oct 27, 2026
2
useAgentChat shows the chat as idle, and Stop does nothing, after a reconnect while a turn is accepted but has not streamed
cloudflare/agents#2741 · opened on Oct 27, 2026
4
Upstream memory and CPU items in pi-ai, pi-durable and computer found while profiling the Pi harness
cloudflare/agents#2736 · opened on Oct 27, 2026
examples/pi harness: cut memory with agents/routing, a minified Latin-1 build and optional git
cloudflare/agents#2735 · opened on Oct 27, 2026
models/pi-ai: the Anthropic SDK chunk costs memory in Workers that only use Workers AI
cloudflare/agents#2733 · opened on Oct 27, 2026
models/pi-ai: importing convertMessages pulls in the whole openai SDK
cloudflare/agents#2732 · opened on Oct 27, 2026
agents dist: about 700 non-Latin-1 characters in published JS double the memory of the modules that contain them
cloudflare/agents#2731 · opened on Oct 27, 2026
2
agents/vite: one non-Latin-1 character makes V8 hold the whole Worker module as a two-byte string
cloudflare/agents#2730 · opened on Oct 27, 2026
1

Navigation Menu

Repository navigation

Issues

Issues

Agent readiness scan aborts homepage GET despite successful HTTP 200 responses

`steering-message` chunks from AI SDK steering are dropped by `applyChunkToParts` and never persisted

Compaction overlay stops applying once the thread passes ~10k rows

agents/streams: list() newest-first tiebreak, async cutover commit, unvalidated read inputs

Think approval turn waits ~30 s (deadman alarm) to recover when the DO is reset during the alarm job

Several models used in examples/docs are being deprecated on Nov 20

`@cloudflare/think` 0.23.0 breaks Workers builds on `just-bash`, which nothing installs.

`Lifecycle.fetch` returns the error's stack to the caller when startup or `onRequest` throws

Streamable HTTP initialize makes an unused getInitializeRequest round trip

worker-bundler: subpath imports to a directory with its own package.json resolve as external

createAI: string catalog models drop Anthropic adaptive thinking and effort providerOptions

ThinkChat sends the full transcript on every message update, and get-messages has no paging

OpenCodeHarness: wait() and prompt() pause 1 s after every run

Think: an interrupted continuation recovered twice is persisted as two assistant rows with the same toolCallId

agents/models/opencode: OpenCode needs to let plugins supply an AI SDK provider on workerd

agents/context: cache the system prompt by default and restore the fluent withContext API

Detached runs: a given-up run whose child was torn down stays outstanding forever, so _cfDetachedReconcileTick re-arms indefinitely

createAI: support turning reasoning off on Workers AI-backed dynamic routes

useAgentChat shows the chat as idle, and Stop does nothing, after a reconnect while a turn is accepted but has not streamed

Upstream memory and CPU items in pi-ai, pi-durable and computer found while profiling the Pi harness

examples/pi harness: cut memory with agents/routing, a minified Latin-1 build and optional git

models/pi-ai: the Anthropic SDK chunk costs memory in Workers that only use Workers AI

models/pi-ai: importing convertMessages pulls in the whole openai SDK

agents dist: about 700 non-Latin-1 characters in published JS double the memory of the modules that contain them

agents/vite: one non-Latin-1 character makes V8 hold the whole Worker module as a two-byte string