Every source GILD analyzed, newest first. Each says who the source is, how it knows, whether it still applies and which findings it counts toward. Open a count to see only its sources.
Source
75 sources in All practitioners, with product documentation Showing all 75 sources
Negative
Approval decision transfers across runs when call ID is reusedGitHub issue
An approval given in one run can authorize the same protected tool call in a different run when the call ID and arguments repeat; the issue is still open.
Positive
Observing AI Agents with OpenTelemetryReddit post
Built an NBA agent with input guardrails, server-backed sessions, and a tool, and picked the SDK for its documentation. OpenTelemetry libraries did not yet record its guardrails.
Mixed
MCPServerStdio defaults to a 5-second session timeout, while an uncached npx or uvx server often takes 10 to 25 seconds to start.
Counts toward, pending a recheck
Negative
When the only stream consumer stopped while handling an event, the run kept calling the model and tools in the background with nobody reading its events.
Counts toward, pending a recheck
Mixed
Tool-calling runs get quadratically slower: completed tool calls are re-fingerprinted every turnGitHub issue
Run time grew quadratically with tool-calling turns: a 300-turn run with a fake model took 2.69 seconds, about 1.2 of them spent re-fingerprinting earlier tool calls.
Counts toward, pending a recheck
Mixed
Follow-up to #5330: recommended guardrail pattern verified on 0.23.1 (results + one output_info papercut)GitHub issue
Confirmed the shared-context guardrail pattern a maintainer suggested, and measured that under the default concurrent setting all 6 tool calls had started before the tripwire fired, in 30 of 30 runs.
Negative
ShellTool: empty ShellResult fabricates stdout "(no output)" in the model-visible shell_call_outputGitHub issue
When a shell command succeeded and printed nothing, the SDK told the model the command's stdout was "(no output)" and stored the same text as the command's output, so the model could quote output that never existed. The issue is closed.
"…the SDK fabricates stdout that the executor never produced:…"
Counts toward, pending a recheck
Negative
aisdk(): parallel tool results become separate tool messages, which breaks GeminiGitHub issue
The AI SDK adapter sends parallel tool results as separate messages, so Gemini rejects the next request after two or more parallel tool calls; the issue is still open.
Mixed
GPT-6 models get no default reasoning settings leading to 400 on /v1/chat/completions with function toolsGitHub issue
Agents on GPT-6 Sol with any function tool failed with a 400 on Chat Completions, because the SDK sent no default reasoning setting for GPT-6 models while the same code worked on gpt-5.6-luna. The issue is closed.
Counts toward, pending a recheck
Mixed
Adding your own trace collector to the OpenAI Agents SDKPost on X
Warns that adding a trace processor leaves OpenAI's exporter running, and only set_trace_processors replaces it.
Negative
RedisSession fails on a Redis Cluster client because two session methods call a pipeline method the cluster client lacks; the issue is still open.
Mixed
Why I combined fetching and extraction in one agent toolReddit post
A lead-generation agent overflowed the model's context when a 540K-character page passed between two tools, so the builder merged steps into compact tools. One directory page then yielded 78 companies.
Counts toward
Positive
First hands-on experience with the OpenAI Agents SDKPost on X
On a first build, found that tracing out of the box made agent handoffs and tool calls much easier to debug.
Negative
Non-streaming Chat Completions run fails when provider usage has null token countsGitHub issue
A non-streaming run against an OpenAI-compatible server failed with a validation error whenever the provider sent a null token count, even though the model had returned a valid answer; streaming runs already handled it. The issue is closed.
Counts toward, pending a recheck
Negative
After the documented end-of-stream signal, the voice pipeline kept waiting about 16.7 minutes with the speech-to-text socket open.
Counts toward, pending a recheck
Negative
Across 42 nested Agent.as_tool calls above the 1,024-token minimum, none returned cached tokens, so the cached-input discount never applied.
Counts toward, pending a recheck
Negative
BackendSpanExporter drops the whole batch on HTTP 429 instead of retrying with Retry-AfterGitHub issue
The tracing exporter dropped a whole batch of spans on an HTTP 429 and skipped retrying after the Retry-After delay.
Counts toward, pending a recheck
Positive
Comment on: OpenAI Agents APIHacker News comment
Has had great success with the SDK, building sandbox, Slack, and knowledge-base integrations separately and limiting what reaches OpenAI. Sees little point in the hosted Agents API, since hosting a Python service is easy.
Mixed
Comment on: OpenAI Agents APIHacker News comment
Uses the SDK because tracking OpenAI's API changes by hand is more trouble than it is worth, and says a portable multi-model agent would skip the OpenAI SDK entirely.
Positive
Comment on: OpenAI Agents APIHacker News comment
Asks why anyone would pick the hosted Agents API over the SDK, which feels better running in a sandbox.
Mixed
Comment on: OpenAI Agents APIHacker News comment
Finds Codex's app-server more flexible than the raw Responses API or the Agents SDK, and says everyone is still working out the right interface.
Mixed
Measured findings - agentic-arenaOpen-source benchmark site
With a scripted model, the SDK sent 856.9 prompt tokens per tool-use item, 1.14 times a hand-rolled loop, and its handoffs added about 251 characters of schema per delegate to every request.
Negative
Responses terminal failures drop provider-reported usageGitHub issue
When the Responses API ended a call incomplete or failed and still reported its usage, the SDK raised the right error but recorded zero requests and zero tokens for it, so failed calls vanished from the run's usage. The issue is closed.
Counts toward, pending a recheck
Negative
AnyLLM and LiteLLM return successful empty output for truncated responsesGitHub issue
The Any-LLM and LiteLLM adapters returned an empty, truncated response marked as success.
Counts toward, pending a recheck
Negative
On main at commit 89c02c8, after #4613, an approval resume could write a tool output without its tool call into the session, after which every later turn of that conversation failed with a 400 and the conversation was permanently dead. In the reproducer 0.21.1 was unaffected and 0.22.0 failed earlier, during the resume itself. The issue is closed.
Counts toward, pending a recheck
Positive
OpenAI Agents SDK: using tool descriptions to route search vs. live-page retrievalReddit post
Routed an agent between WebSearchTool and a custom page fetcher by writing distinct tool descriptions, after which it searched first and fetched pages when it needed their content.
Mixed
Comment on: Launch HN: Hoplite (YC S26)Hacker News comment
Building cloud coding agents from the ground up on the SDK inside an existing in-house environment, and says the work to turn it into an agent is not small.
Mixed
Claude Agent SDK vs OpenAI Agents SDK: I Built the Same Agent in BothPersonal newsletter (Substack)
Built one lead research agent in both SDKs. The OpenAI version cost US$0.08 a run with GPT-5, took 22 seconds and 125 lines, and hallucinated where the Claude version refused; the author finds the SDK built around OpenAI.
Positive
Comment on: Ask HK: How are you building AI apps today?Hacker News comment
Finds handoffs and multi-agent orchestration intuitive, and values the sessions, chat history, and boilerplate the SDK adds over the Responses API.
Positive
Comment on: Ask HK: How are you building AI apps today?Hacker News comment
Moved every older Completions and Responses API project to the SDK, finds it stable, and uses it with other inference services and local models in LM Studio.
Mixed
OpenAI Agents SDK Sandboxes: Which one should you choose?Hacker News comment
Compared the SDK's supported sandboxes on isolation, persistence, pricing, GPU support, and data residency, because the tradeoffs were not clear from the docs.
Not counted
Superserve sells sandbox infrastructure for agents built on the SDK, so its founder has a commercial stake. Kept in the record; it takes part in no figure.
Mixed
OpenAI Agents SDK Sandboxes: Which Provider Should You Actually Use?Company blog
OpenAI ships sandbox clients for seven hosted options plus two local ones without ranking them. Cold starts can dominate short tasks, and pricing is published unevenly across them.
Not counted
Superserve sells sandbox infrastructure for agents built on the SDK, and the post says it has a stake. Kept in the record; it takes part in no figure.
Mixed
Comment on: Claude Managed AgentsHacker News comment
Largely happy with the SDK's handoffs, tools, guardrails, and structured output, but found building infrastructure and observability that scale the bigger pain, which explains the pull of managed agents.
Positive
Comment on: Ask HN: Learning resources for building AI agents?Hacker News comment
Advises starting minimal with the OpenAI Agents SDK and calls most other frameworks overengineered.
Positive
Show HN: Runtime Use, run agents in any sandboxHacker News post
Lark's team open-sourced the sandbox runtime it built for its test-generating agents, which by default runs an Agents SDK agent with shell and web search tools and can switch to a Claude Code agent.
"By default it runs an OpenAI Agents SDK agent with shell + web search tools."
No opinion stated
Comment on: OpenAI's unit economicsHacker News comment
Reads the Agents SDK, the Assistants API, and ChatGPT Apps as OpenAI's lock-in layer on top of a commodity Responses API.
Negative
With a compaction session on a streamed run, the stream stopped showing output and did not end until compaction finished, adding a wait after the answer.
Counts toward, pending a recheck
Negative
Memory leak: Agent references not released between turnsGitHub issue
Agent objects stayed in memory across runs in long workflows because run items kept strong references to them.
Counts toward, pending a recheck
Positive
Comment on: Getting a Gemini API key is an exercise in frustrationHacker News comment
Calls Google's ADK over-abstracted and points to smolagents, Pydantic AI, or the simpler OpenAI Agents SDK for a lightweight framework.
No opinion stated
How Agentic Search Unlocks Legal Research Intelligence at HarveyCompany engineering blog
Harvey runs offline evaluations of its agentic legal search in LangSmith on the Agents SDK's OpenTelemetry traces, and the eval data lifted its tool selection precision from near zero to 0.8 to 0.9.
"Centralized evaluation infrastructure that we built using the OpenAI Agent SDK's OpenTelemetry traces enables us to run offline evaluations in LangSmith. We track metrics like hallucination, tool recall, retrieval recall, formatting, and answer quality."
Negative
Traces are silently dropped in long-running workersGitHub issue
Traces created inside long-running background workers never reached OpenAI's Traces dashboard, with no error.
Counts toward, pending a recheck
Mixed
Resilience mechanism Openai/ LiteLLMGitHub issue
Could not get retries to work after moving an agent to Claude through LitellmModel, and found no documentation on resilience settings.
Counts toward, pending a recheck
Negative
Item with id rs_ not foundGitHub issue
A triage agent on GPT-5 with minimal reasoning stored an empty reasoning item in its SQLite session, and once control came back to it after two handoffs every new message failed with a 404 for that item. The issue is closed.
Counts toward, pending a recheck
Positive
Comment on: Inkeep's agent builder launchHacker News comment
Asks why a team would take Inkeep, which is under the Elastic License 2.0, over the MIT-licensed Agents SDK.
Mixed
Comment on: Ask HN: Do you roll your own agent or use a framework?Hacker News comment
Used LangChain, LangGraph, and the Agents SDK extensively, then built an in-house async framework that the team finds much cleaner.
Not counted
Its author maintains graphai, Aurelio Labs' agent framework, which competes with the Agents SDK. Kept in the record; it takes part in no figure.
Negative
Unacceptable latency with 4.1GitHub issue
After a tool call, GPT-4.1 often took up to 30 seconds to answer with few tokens, and the issue questions whether businesses can adopt the SDK at that speed.
Counts toward, pending a recheck
Positive
Comment on: Open SWE by LangChainHacker News comment
Names OpenAI's Agents SDK and Pydantic AI as solid alternatives to LangGraph for agent and workflow orchestration.
Negative
Why have agents (or GPT 4.1) slowed down so much?GitHub issue
With no code change, one agent call on GPT-4.1 went from under 4 seconds to over 11 seconds, all of it spent in the Responses API.
Counts toward, pending a recheck
Mixed
Comment on: OpenAI-agents-Redis: Native OpenAI Agents SDK session management using RedisHacker News comment
Built a Redis session store for the SDK because the built-in SQLite sessions could not scale across instances or survive container restarts.
Negative
Agent run with previous_response_id fails - No tool output found for function callGitHub issue
Runs chained with previous_response_id failed sporadically after tool calls with a missing tool output error; the issue drew 24 comments before it closed.
Counts toward, pending a recheck
Mixed
Comment on: Show HN: My LLM CLI tool can run tools now, from Python code or pluginsHacker News comment
Found tool calls hit-and-miss in a calendar assistant on the SDK, with GPT-4.1 claiming it had added events without calling the tool.
Counts toward, pending a recheck
Negative
LiteLLM + Gemini 2.5 Pro: cached_tokens=None crashes Agents SDK with Pydantic int-validation errorGitHub issue
Runs with Gemini 2.5 Pro through LiteLLM crashed during cost calculation because the model returned no cached-token count.
Counts toward, pending a recheck
Positive
AI VC Simulations Built on OpenAI's Agents SDKHacker News comment
Octagon AI open-sourced an MCP server that runs VC simulations on the SDK, citing multi-agent reasoning, context handoff, and tool use out of the box.
Positive
Comment on: Show HN: RowboatHacker News comment
Rowboat's agents runtime and copilot are Python apps built on the SDK, beside a Next.js dashboard.
Counts toward
No opinion stated
Optimize latency for parallel agent runs with streamingGitHub issue
Adhish Thite's three parallel translation agents took about 10 seconds before the picker agent could start, so the streaming UI showed nothing until then, and the wait compounds in longer chains.
"While this approach provides quality benefits, it introduces significant latency in the user experience, especially when streaming responses."
Not counted
Its author's ties to the vendor have not been checked. His GitHub profile lists @google as his company and GenAI @ Elastic in its bio. Kept in the record; it takes part in no figure.
Negative
Long response times when using multiple agents with agents_as_toolsGitHub issue
Single agent calls in a seven-agent orchestrator took about 90 seconds in traces, most of it in the Responses API.
Counts toward, pending a recheck
No opinion stated
Comment on: Open Deep ResearchHacker News comment
Built an open-source deep research agent on the newly released SDK that plans a topic, splits it, and researches sub-topics in parallel. The limits the author lists concern the models' long-context output.
Counts toward
Mixed
Duplicate traces through OpenAI agents SDKGitHub issue
Each Agents SDK trace appeared three times in LangSmith with different ids, which made the team's dashboards untrustworthy. A suggested pre-release then broke handoffs and was reverted.
Counts toward, pending a recheck
No opinion stated
feat: python openai agents sdk supportGitHub pull request
Adds an Agents SDK extension to Coinbase's AgentKit in Python, tested with a chatbot that reported its wallet address and requested testnet funds; AgentKit's repository now ships three example chatbots on the SDK.
Pages from OpenAI, used to check facts for every group.
Describes the SDK as a lightweight package with very few abstractions and a production-ready upgrade of Swarm. Its primitives are agents, agents as tools and handoffs, guardrails, sessions, human-in-the-loop approvals, MCP tool calling, realtime voice agents, sandbox agents, and tracing.
With no model set, agents use gpt-5.6-luna with reasoning effort none and low verbosity. OPENAI_DEFAULT_MODEL or RunConfig changes the default. LiteLLM and Any-LLM support is best-effort beta, and many other models lack structured outputs, multimodal input, and hosted search.
States that the default is currently gpt-5.6-luna with reasoning effort none and low text verbosity.
The SDK works with OpenAI models through the Responses or Chat Completions API, and other models come in through the Vercel AI SDK adapter.
Tracing is on by default and exports batches to the OpenAI backend. It is unavailable to organizations under Zero Data Retention. add_trace_processor() keeps the OpenAI exporter, set_trace_processors() replaces it, and flush_traces() exports buffered traces in long-running workers.
Documents session memory backed by SQLite, Redis, SQLAlchemy, MongoDB, Dapr, and OpenAI Conversations, among other stores.
Guardrails run in parallel with the agent by default, so a tripwire can fire after the agent has spent tokens or run tools; blocking mode runs them first.
Tools declare when they need approval, results surface pending approvals as interruptions, and RunState serializes a paused run so it can resume after a decision.
The SDK uses 0.Y.Z versions, and a minor version may include breaking changes to public interfaces that are not marked beta.
Lists US$0.20 per million input tokens and US$1.20 per million output tokens, a 1,050,000-token context window, and 128,000 max output tokens. It is not supported on the Free usage tier and starts at 5,000 RPM and 2,000,000 TPM on Build.
Spend alerts send notifications without stopping API traffic, and hard spend limits can stop affected requests when tracked spend reaches the configured amount. Organization and project limits can both apply.
April 15, 2026: the SDK gains a harness and sandbox agents, first in Python, with built-in support for Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, and Vercel sandboxes. Snapshotting and rehydration restore an agent's state in a fresh container when the original fails or expires, and the capabilities bill at standard API prices for tokens and tool use.
OpenAI is winding down Agent Builder and Evals, which leave the OpenAI platform on November 30, 2026. For workflows that should continue as code, OpenAI recommends the Agents SDK.
September 10, 2026: OpenAI's hosted Agents API enters public beta with the harness that powers Codex, and it charges no fee beyond the tokens and tools agents use.
Lists elevated errors across the API including the Agents API on October 6, elevated gpt-5.6-luna API error rates on October 21, and elevated Responses API errors on October 22, 2026.
MIT-licensed repository with 30,355 stars on October 28, 2026; v0.24.0 was released on October 15, 2026.
Source index
Every source GILD read, with the original address and the context used to count it. Open a finding to filter the full cards above. On smaller screens the index lists each source once by name, cohort and date; its full record above carries the address, context and findings.
| Source | Original URL | Author and context | Who | Published | Review date | Applies | Findings |
|---|---|---|---|---|---|---|---|
| Approval decision transfers across runs when call ID is reusedID gh-py-5413 | Developer Communities | github.com/ | Tardfyou, GitHub user filing an issueself-described: Approval-gated tools, tested on 0.23.1 | Individual builder | Published Oct 24, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Observing AI Agents with OpenTelemetryID rd-1xcbmxo | Developer Communities | www.reddit.com/ | silksong_when, Reddit poster who works at SigNozself-described: NBA question-answering agent instrumented with OpenTelemetry | Individual builder | Published Oct 22, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| docs(mcp): clarify client_session_timeout_seconds default (5s), npx/cli cold-start timeout, and document missing constructor optionsID gh-py-5405 | Developer Communities | github.com/ | MuneebMalik244535, GitHub user filing an issueself-described: MCP servers started with npx | Individual builder | Published Oct 21, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Streamed run keeps calling the model and tools after its only stream_events() consumer stops outside the queue waitID gh-py-5386 | Developer Communities | github.com/ | sushantsimpplr, GitHub user filing an issueself-described: Streamed runs whose consumer stops early | Individual builder | Published Oct 17, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Tool-calling runs get quadratically slower: completed tool calls are re-fingerprinted every turnID gh-py-5382 | Developer Communities | github.com/ | deadstrobe5, GitHub user filing an issueself-described: Long tool-calling runs | Individual builder | Published Oct 16, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Follow-up to #5330: recommended guardrail pattern verified on 0.23.1 (results + one output_info papercut)ID gh-py-5351 | Developer Communities | github.com/ | PHM-Audit, GitHub user filing an issueself-described: Guardrails that block out-of-plan tool calls, on 0.23.1 | Individual builder | Published Oct 12, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| ShellTool: empty ShellResult fabricates stdout "(no output)" in the model-visible shell_call_outputID sf-5 | Developer Communities | github.com/ | BlueX888, GitHub profile lists Beijing University of Technologyself-described: Shell tool runs for commands that succeed and print nothing | Individual builder | Published Oct 11, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| aisdk(): parallel tool results become separate tool messages, which breaks GeminiID gh-js-2048 | Developer Communities | github.com/ | igorbereza, GitHub user filing an issueself-described: Gemini through the TypeScript AI SDK adapter | Individual builder | Published Oct 8, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| GPT-6 models get no default reasoning settings leading to 400 on /v1/chat/completions with function toolsID gh-py-5336 | Developer Communities | github.com/ | Loïc Minaudier, GitHub profile lists Datadogself-described: An agent on GPT-6 Sol with function tools, called through Chat Completions behind an OpenAI-compatible gateway | Product or platform team | Published Oct 8, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Adding your own trace collector to the OpenAI Agents SDKID x-gocronk | Developer Communities | x.com/ | Andrew (gocronk), heavy AI user on Xself-described: Keeping agent traces in the team's own system | Individual builder | Published Oct 7, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| RedisSession fails on Redis Cluster: add_items/clear_session raise AttributeError ('ClusterPipeline' has no 'raise_first_error')ID gh-py-5322 | Developer Communities | github.com/ | sushantsimpplr, GitHub user filing an issueself-described: Sessions stored on a Redis Cluster | Individual builder | Published Oct 6, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Why I combined fetching and extraction in one agent toolID rd-1wywdni | Developer Communities | www.reddit.com/ | OkShirt9372, Reddit poster in r/AI_Agentsself-described: Lead-generation agent built with the SDK and ZenRows | Individual builder | Published Oct 6, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| First hands-on experience with the OpenAI Agents SDKID x-chetanwadhwa03 | Developer Communities | x.com/ | Chetan Wadhwa, developer on Xself-described: First hands-on build with the SDK | Individual builder | Published Oct 3, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Non-streaming Chat Completions run fails when provider usage has null token countsID sf-4 | Developer Communities | github.com/ | Cheolhee Lee, AI full-stack developer at a B2B AI software companyself-described: Agents on an OpenAI-compatible server through the Chat Completions model | Individual builder | Published Sep 28, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Voice: add_audio(None) doesn't end the STT session, so the pipeline waits ~1000s for EVENT_INACTIVITY_TIMEOUTID gh-py-5191 | Developer Communities | github.com/ | kosesena, GitHub user filing an issueself-described: Voice pipeline with streaming speech-to-text | Individual builder | Published Sep 27, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Agent.as_tool() nested subagent runs never hit prompt cache, even with byte-identical instructions/tools/prompt_cache_keyID gh-py-5085 | Developer Communities | github.com/ | prajwal-bik-ai, GitHub user filing an issueself-described: Top-level agent with nested subagents attached as tools | Individual builder | Published Sep 18, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| BackendSpanExporter drops the whole batch on HTTP 429 instead of retrying with Retry-AfterID gh-py-5023 | Developer Communities | github.com/ | coderdailyone, GitHub user filing an issueself-described: Trace export under rate limiting | Individual builder | Published Sep 14, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Comment on: OpenAI Agents APIID hn-49653842 | Developer Communities | news.ycombinator.com/ | baalimago, Hacker News commenterself-described: Company agent with sandbox, Slack, and knowledge-base integrations | Individual builder | Published Sep 11, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: OpenAI Agents APIID hn-49654265 | Developer Communities | news.ycombinator.com/ | baalimago, Hacker News commenterself-described: Company agent with sandbox, Slack, and knowledge-base integrations | Individual builder | Published Sep 11, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: OpenAI Agents APIID hn-49650188 | Developer Communities | news.ycombinator.com/ | maxdo, Hacker News commenterself-described: Discussion of OpenAI's hosted Agents API launch | Individual builder | Published Sep 10, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: OpenAI Agents APIID hn-49651406 | Developer Communities | news.ycombinator.com/ | jumploops, Hacker News commenterself-described: Discussion of OpenAI's hosted Agents API launch | Individual builder | Published Sep 10, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Measured findings - agentic-arenaID blog-agentic-arena | Engineering Blogs & Benchmarks | code-with-rashid.github.io/ | code-with-rashid, author of the open-source agentic-arena benchmarkself-described: Scripted-model comparison of agent frameworks against a hand-rolled loop | Individual builder | Published Sep 8, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Responses terminal failures drop provider-reported usageID sf-3 | Developer Communities | github.com/ | RongJie G (CorgiBoyG), GitHub profile lists Xidian Universityself-described: Token accounting on runs whose Responses call ends incomplete or failed | Individual builder | Published Sep 7, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| AnyLLM and LiteLLM return successful empty output for truncated responsesID gh-py-4885 | Developer Communities | github.com/ | CoralGarden52, GitHub user filing an issueself-described: Chat Completions models through the Any-LLM and LiteLLM adapters | Individual builder | Published Sep 6, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Deferred interrupted-turn session items are never persisted when an approval resume goes through next_step_run_againID sf-2 | Developer Communities | github.com/ | Julio de la Calle (dixso), GitHub user filing an issueself-described: An agent with output guardrails and a tool that needs approval, resumed after the decision | Individual builder | Published Sep 2, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| OpenAI Agents SDK: using tool descriptions to route search vs. live-page retrievalID rd-1w3al0f | Developer Communities | www.reddit.com/ | InsideDebt6345, Reddit poster in r/AI_Agentsself-described: Two-tool agent with web search and a page fetcher | Individual builder | Published Aug 31, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: Launch HN: Hoplite (YC S26)ID hn-49160727 | Developer Communities | news.ycombinator.com/ | asdev, Hacker News commenterself-described: Team with its own in-house cloud development environment | Small product team | Published Aug 3, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Claude Agent SDK vs OpenAI Agents SDK: I Built the Same Agent in BothID blog-elshadk | Engineering Blogs & Benchmarks | elshadk.substack.com/ | Elshad Karimov, builds production agents for clientsself-described: Same lead research agent built in the Claude Agent SDK and the OpenAI Agents SDK | Individual builder | Published Jun 13, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: Ask HK: How are you building AI apps today?ID hn-48356738 | Developer Communities | news.ycombinator.com/ | sibidharan, Hacker News commenterself-described: EdTech apps with AI learning and evaluation tools | Individual builder | Published Jun 1, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: Ask HK: How are you building AI apps today?ID hn-48326951 | Developer Communities | news.ycombinator.com/ | sibidharan, Hacker News commenterself-described: EdTech apps with AI learning and evaluation tools | Individual builder | Published May 29, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| OpenAI Agents SDK Sandboxes: Which one should you choose?ID hn-48211199 | Developer Communities | news.ycombinator.com/ | pavitrabhalla, founder of Superserve, a sandbox companyself-described: Compared the sandbox options the SDK supports | Technical leadership | Published May 20, 2026 | Assessment reviewed Nov 1, 2026 | Current | Not counted. Superserve sells sandbox infrastructure for agents built on the SDK, so its founder has a commercial stake. |
| OpenAI Agents SDK Sandboxes: Which Provider Should You Actually Use?ID blog-superserve | Engineering Blogs & Benchmarks | www.superserve.ai/ | Pavitra Bhalla, founder of Superserve, a sandbox companyself-described: Comparison of the sandbox options the SDK supports | Technical leadership | Published May 19, 2026 | Assessment reviewed Nov 1, 2026 | Current | Not counted. Superserve sells sandbox infrastructure for agents built on the SDK, and the post says it has a stake. |
| Comment on: Claude Managed AgentsID hn-47707530 | Developer Communities | news.ycombinator.com/ | suncemoje, Hacker News commenterself-described: Has run agents on the SDK for a while | Individual builder | Published Apr 9, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: Ask HN: Learning resources for building AI agents?ID hn-47647875 | Developer Communities | news.ycombinator.com/ | stochtinkerer, Hacker News commenterself-described: Advice to a developer starting to build agents | Individual builder | Published Apr 5, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Show HN: Runtime Use, run agents in any sandboxID sf-8 | Developer Communities | news.ycombinator.com/ | Vijit and Jack, builders of Runtime Use and Lark (Show HN authors)self-described: Lark's agents generate end-to-end tests inside sandboxes | Small product team | Published Mar 18, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: OpenAI's unit economicsID hn-46809121 | Developer Communities | news.ycombinator.com/ | senko, Hacker News commenterself-described: OpenAI's product strategy and API lock-in | Individual builder | Published Jan 29, 2026 | Assessment reviewed Nov 1, 2026 | Current | |
| Compaction introduces post-output latency: stream_events() doesn't terminate after streaming output completesID gh-py-2343 | Developer Communities | github.com/ | Filimoa, GitHub user filing an issueself-described: Streamed runs with a compaction session | Individual builder | Published Jan 20, 2026 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Memory leak: Agent references not released between turnsID gh-py-2217 | Developer Communities | github.com/ | ruskaruma, GitHub user filing an issueself-described: Long-running workflows that run an agent in a loop | Individual builder | Published Dec 20, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Comment on: Getting a Gemini API key is an exercise in frustrationID hn-46308757 | Developer Communities | news.ycombinator.com/ | quleap, Hacker News commenterself-described: Comparing Google's ADK with lighter agent frameworks | Individual builder | Published Dec 18, 2025 | Assessment reviewed Nov 1, 2026 | Current | |
| How Agentic Search Unlocks Legal Research Intelligence at HarveyID sf-12 | Engineering Blogs & Benchmarks | www.harvey.ai/ | Lysia Li, Varun Nair, Aaron Stern, Pablo Felgueres, Calvin Qi and Philip Cerles, Harvey engineeringself-described: Evaluating Harvey's agentic legal search | Product or platform team | Published Dec 10, 2025 | Assessment reviewed Nov 1, 2026 | Current | Listed, no finding position |
| Traces are silently dropped in long-running workersID gh-py-2135 | Developer Communities | github.com/ | kiansdi, GitHub user filing an issueself-described: Agents run in Celery and FastAPI background workers | Individual builder | Published Nov 29, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Resilience mechanism Openai/ LiteLLMID gh-py-2050 | Developer Communities | github.com/ | dast-draftwise, GitHub user filing an issueself-described: Moving an agent from OpenAI models to Claude through LiteLLM | Individual builder | Published Nov 6, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
Item with id rs_ not foundID sf-1 | Developer Communities | github.com/ | Dhanush Kumar Shankar (shankar-dh), GitHub user filing an issueself-described: Three agents on GPT-5 with handoffs and the default SQLiteSession | Individual builder | Published Oct 31, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Comment on: Inkeep's agent builder launchID sf-7 | Developer Communities | news.ycombinator.com/ | d0100, Hacker News commenterself-described: Choosing between Inkeep and the Agents SDK for an agent product | Individual builder | Published Oct 16, 2025 | Assessment reviewed Nov 1, 2026 | Current | Listed, no finding position |
| Comment on: Ask HN: Do you roll your own agent or use a framework?ID hn-45512772 | Developer Communities | news.ycombinator.com/ | jamesbriggs, Hacker News commenterself-described: Team that moved from frameworks to its own async framework | Small product team | Published Oct 8, 2025 | Assessment reviewed Nov 1, 2026 | Current | Not counted. Its author maintains graphai, Aurelio Labs' agent framework, which competes with the Agents SDK. |
| Unacceptable latency with 4.1ID gh-py-1664 | Developer Communities | github.com/ | sibblegp, GitHub user filing an issueself-described: Client apps that run agents on GPT-4.1 | Small product team | Published Sep 4, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Comment on: Open SWE by LangChainID hn-44840626 | Developer Communities | news.ycombinator.com/ | NeutralCrane, Hacker News commenterself-described: Comparing LangChain's products with alternatives | Individual builder | Published Aug 8, 2025 | Assessment reviewed Nov 1, 2026 | Current | |
| Why have agents (or GPT 4.1) slowed down so much?ID gh-py-1249 | Developer Communities | github.com/ | sibblegp, GitHub user filing an issueself-described: Client apps that run agents on GPT-4.1 | Small product team | Published Jul 25, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Comment on: OpenAI-agents-Redis: Native OpenAI Agents SDK session management using RedisID hn-44644196 | Developer Communities | news.ycombinator.com/ | j0selit0, Hacker News commenter (author of openai-agents-redis)self-described: Production app whose agents run on several instances | Individual builder | Published Jul 22, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Agent run with previous_response_id fails - No tool output found for function callID gh-py-1061 | Developer Communities | github.com/ | markns, GitHub user filing an issueself-described: Multi-turn runs chained with previous_response_id | Individual builder | Published Jul 11, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Comment on: Show HN: My LLM CLI tool can run tools now, from Python code or pluginsID hn-44111402 | Developer Communities | news.ycombinator.com/ | stavros, Hacker News commenterself-described: Personal assistant with access to a calendar | Individual builder | Published May 27, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| LiteLLM + Gemini 2.5 Pro: cached_tokens=None crashes Agents SDK with Pydantic int-validation errorID gh-py-758 | Developer Communities | github.com/ | ruidazeng, GitHub user filing an issueself-described: Gemini 2.5 Pro through the LiteLLM adapter | Individual builder | Published May 26, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| AI VC Simulations Built on OpenAI's Agents SDKID hn-43800384 | Developer Communities | news.ycombinator.com/ | ag_21, Octagon AI (Show HN author)self-described: Octagon VC Agents, an open-source MCP server | Small product team | Published Apr 26, 2025 | Assessment reviewed Nov 1, 2026 | Current | |
| Comment on: Show HN: RowboatID hn-43765732 | Developer Communities | news.ycombinator.com/ | ramnique, Rowboat co-founder (Show HN author)self-described: Rowboat, an open-source IDE for multi-agent systems | Technical leadership | Published Apr 22, 2025 | Assessment reviewed Nov 1, 2026 | Current | |
| Optimize latency for parallel agent runs with streamingID sf-6 | Developer Communities | github.com/ | Adhish Thite, GenAI engineer (GitHub profile)self-described: Three translation agents run in parallel, then a picker agent, behind a streaming UI | Individual builder | Published Apr 14, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | Not counted. Its author's ties to the vendor have not been checked. His GitHub profile lists @google as his company and GenAI @ Elastic in its bio. |
| Long response times when using multiple agents with agents_as_toolsID gh-py-421 | Developer Communities | github.com/ | PaulSGit1, GitHub user filing an issueself-described: Orchestrator on GPT-4.5 with seven agents as tools | Individual builder | Published Apr 2, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| Comment on: Open Deep ResearchID hn-43553242 | Developer Communities | news.ycombinator.com/ | hackfather, Hacker News commenterself-described: Open-source deep research agent built on the SDK at launch | Individual builder | Published Apr 2, 2025 | Assessment reviewed Nov 1, 2026 | Current | |
| Duplicate traces through OpenAI agents SDKID gh-ls-1619 | Developer Communities | github.com/ | Karolis Mariunas (kmariunas), Developer (issue author)self-described: Agents SDK app with handoffs, traced to LangSmith | Individual builder | Published Mar 31, 2025 | Assessment reviewed Nov 1, 2026 | Needs a recheck | |
| feat: python openai agents sdk supportID sf-11 | Developer Communities | github.com/ | Carson Roscoe (CarsonRoscoe), contributor to Coinbase Developer Platform's AgentKitself-described: AgentKit, Coinbase Developer Platform's toolkit for giving agents crypto wallets | Product or platform team | Published Mar 11, 2025 | Assessment reviewed Nov 1, 2026 | Current | Listed, no finding position |
| OpenAI Agents SDK (Python) overviewID pd-01 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Agents SDK (Python): ModelsID pd-02 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Agents SDK (TypeScript): ModelsID pd-03 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Agents SDK (TypeScript): AI SDK extensionID pd-04 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Agents SDK (Python): TracingID pd-05 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Agents SDK (Python): SessionsID pd-06 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Agents SDK (Python): GuardrailsID pd-07 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Agents SDK (Python): Human in the loopID pd-08 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Agents SDK (Python): Release processID pd-09 | Product Documentation | openai.github.io/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| gpt-5.6-luna model pageID pd-10 | Product Documentation | developers.openai.com/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| OpenAI API usage and spend limitsID pd-11 | Product Documentation | help.openai.com/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| OpenAI: The next evolution of the Agents SDKID pd-12 | Product Documentation | openai.com/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| OpenAI: Introducing AgentKit (update of June 3, 2026)ID pd-13 | Product Documentation | openai.com/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| Introducing the Agents APIID pd-14 | Product Documentation | openai.com/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| OpenAI status historyID pd-15 | Product Documentation | status.openai.com/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
| openai-agents-python on GitHubID pd-16 | Product Documentation | github.com/ | OpenAIOfficial OpenAI Agents SDK documentation and OpenAI platform pages | Not counted (documentation) | No date stated | Documentation checked Oct 28, 2026 | Current | Verification only |
