Firecrawl做開發索引
Introducing the Firecrawl Developer Index, built for supercharging coding agents.Read the announcement →Developer Search IndexTechnical docs searchfor coding agents70M+ docs, READMEs, issues, and merged pull requests in one index, with the highest recall of any coding-specific index.
$npx -y firecrawl-cli@latest setup developer-indexCopy skillView docs$ firecrawl developer "why is my retry backoff not firing on 429" --limit 10$ firecrawl developer "why is my retry backoff not firing on 429" --limit 10issue:sidekiq/sidekiq#1471Retries are not applied to 429 responsesThe client treats 429 as a terminal status, so backoff never runs.
pullrequest:aws/aws-sdk-js-v3#2384fix(retry): honor Retry-After on 429Adds 429 to the throttling error list so the standard retry applies.doc:docs.stripe.com/rate-limitsRate limitsHandle 429 responses with an exponential backoff before each retry.
Primary sources70M+Primary sourcesMost sources refreshedDailyMost sources refreshedPlus an MCP server1 APIPlus an MCP serverTo get startedNo API keyTo get started//Inside the index//The libraries your agentsask about70M+ technical docs[ NOT EXHAUSTIVE ]Next.jsReactVueSvelteAngularAstroRemixNuxtNode.
jsDenoBunViteTailwind CSSPlaywrightExpoNext.jsReactVueSvelteAngularAstroRemixNuxtNode.jsDenoBunViteTailwind CSSPlaywrightExpoTypeScriptPythonRustGoSwiftKotlin.NETFlutterDjangoFastAPIFlaskRailsLaravelSpringtRPCTypeScriptPythonRustGoSwiftKotlin.
NETFlutterDjangoFastAPIFlaskRailsLaravelSpringtRPCPostgreSQLRedisMongoDBSQLiteSupabasePrismaKubernetesDockerTerraformCloudflareKafkaGraphQLPyTorchOpenAIPostgreSQLRedisMongoDBSQLiteSupabasePrismaKubernetesDockerTerraformCloudflareKafkaGraphQLPyTorchOpenAISources70M+TypesREADMEs · issues · PRs · docs · specsBrowse the docs[ 01 / 04 ]·Features//Built for coding agents//The sources your agent needs,ranked and ready70M+ artifactsPrimary sources, not web pages.
READMEs, issues, merged pull requests, docs, and OpenAPI specs.readme:vercel/next.jsissue:sidekiq/sidekiq#1471pullrequest:pydantic/pydantic#8291doc:docs.stripe.com/webhooksSkill, CLI, MCP, and APIOne line into any agent.Claude Code, Cursor, and Codex search the index on their own.
terminal$npx -y firecrawl-cli@latest \ setup developer-index✓firecrawl-developer-index readyWired intoClaude CodeCursorCodexMCPMatched passagesThe answer, not just the link.Matched passages come back as markdown, so there is no second scrape.issue:sidekiq/sidekiq#1471github.
com/sidekiq/sidekiqmatched passagemarkdownThe client treats 429 as terminal, so theSemantic retrievalAsk the way you'd ask a person.Paste an error message, get the thread that solved it.
why is my retry backoff not firing on 429Top matchesissue:sidekiq/sidekiq#1471#1pullrequest:aws/aws-sdk-js-v3#2384#2doc:docs.stripe.com/rate-limits#3Continuous refreshToday's thread, not last quarter's.Most sources land within a day of publishing.index refreshlivegithub.com/vercel/next.
jsjust nowdocs.stripe.com1 hr agogithub.com/pydantic/pydantic18 min agodocs.astral.sh/uv4 min agoRich filtersScope it down to one repo.Filter by type, repo, language, topic, license, or stars.narrow the indextypes:issuerepos:vercel/next.jslanguage:Rustminstars:1000license:MITsources:docs.stripe.
com70M+→↓26.2Mresults[ 02 / 04 ]·Benchmarks//DevDex, an open benchmark//The highest recall of anycoding-specific index1,179 real developer queries.Scored deterministically, no model judge.Recall@10, all tracks combined[ HIGHER IS BETTER ]Firecrawl Developer Index0.63Firecrawl Search0.58Parallel0.
57Mintlify0.54Exa0.54Context7docs only0.17Native web search0.45Native web search is the control.With no search tool, the same model scores 0.01.Repository discoveryTrails0.76Find the repo behind a capability, without its name.Best on this track: Parallel·0.82Issues & pull requestsLeads0.
66Find the bug report and the PR that fixed it.Next best: Parallel·0.63Documentation lookupTied0.47Find the page that answers a how-to question.Statistically tied with Context7·0.47Queries1,179Tracks3DriverClaude Opus 4.
8JudgeNoneRun it yourselfTrusted by 150,000+ companies of all sizesRead storyRead storyRead storyRead storyRead storyRead storyRead storyRead story[ 03 / 04 ]·For AI agents//For AI agents//Built into the AI toolsyou already useUse Firecrawl from Claude Code Cursor Windsurf and the agent harnesses your team already runs.
One command.Connect your agent to Firecrawl via our Skills/CLI or MCP.SkillMCPnpx -y firecrawl-cli@latest init --all --browserView the docsAgent onboarding.Are you an AI agent?Fetch this skill to get an API key and start building.cURLcurl -s https://firecrawl.dev/agent-onboarding/SKILL.
mdView the skillcURLCLISkillPythonNode.jsCopycurl -X POST \ https://api.firecrawl.
dev/v2/search/developer \ -H "Content-Type: application/json" \ -d '{ "query": "retry backoff not firing on 429", "k": 10, "types": ["issue", "pullrequest"] }'npm install -g firecrawl-cli firecrawl developer "retry backoff not firing on 429" # Or pipe it into your own tooling firecrawl developer "add pydantic with uv" --jsonnpx -y firecrawl-cli@latest setup developer-index # That is the whole setup.
Your agent can now # search 70M+ READMEs, issues, PRs, and docs: # # > find the PR that fixed the 429 retry bugfrom firecrawl import Firecrawl firecrawl = Firecrawl(apikey="fc-YOUR-API-KEY") result = firecrawl.
search( "retry backoff not firing on 429", categories=["developer"], limit=10, ) for item in result.developer or []: print(item.url, item.title)import { Firecrawl } from "firecrawl"; const firecrawl = new Firecrawl({ apiKey: "fc-YOUR-API-KEY", }); const result = await firecrawl.
search( "retry backoff not firing on 429", { categories: ["developer"], limit: 10 }, ); for (const item of result.developer ??[]) { console.log(item.url, item.title); }No API key.Send one for higher rate limits.
[ 04 / 04 ]·FAQ//FAQ//Frequently asked questionsGeneralWhat is the Firecrawl Developer Index?Why not just use web search?What exactly is in the index?How fresh is the data?Using itHow do I call the Developer Index?Do I need an API key?Is there a Python or Node SDK method?
How do I give my coding agent access?What can I filter on?Benchmarks & pricingWhat is DevDex?Can I reproduce the results?How much does it cost?Is it generally available?
Related
相關文章
Google 開源 Harness²:遞迴代理程式框架,無需新模型也能提升效能
Google Cloud AI Research 開源了 Harness²,這是一個遞迴代理程式框架,透過改良凍結模型周圍的程式碼與檔案來增強效能。對開發者而言,重點很簡單:你的代理程式無需更強大的模型也能進步。重點摘要:規模為框架而非模型,訓練參數為0,包含8個可編輯的框架元件,版本0.1.0。執行環境為Linux、Python 3.12或3.13搭配uv,模型呼叫則透過Vertex AI。GPU需求未公開。效能方面:一次改良步驟即可在3個基準測試中,提升所有4個模型與框架組合的表現。最佳成果:GLM-5搭配OpenCode在WorkBuddy-Bench上從64.2提升至75.7(+11.5點);最差表現:GLM-5搭配Codex在JobBench上僅提升1.2點(23.6至24.8)。什麼是Harness²?它是Google研究推出的即插即用框架。
如何利用NVIDIA結構化資料模型進行零訓練表格預測
In this tutorial, we implement Structured Data Models, NVIDIA’s new library that puts several pretrained foundation models for tables behind one PyTorch interface, and use it to make predictions without training.
認識 Kyutai REKI:2.4B VLA 模型為機器人操控帶來高效記憶
Kyutai 與 Inria 及 École des Ponts 合作,推出了 REKI——一個用於機器人操控的 2.4B 參數視覺-語言-動作模型(VLA)。該模型僅使用公開數據進行預訓練,並配備了輕量級的近期記憶功能。在三個分佈外基準測試中,僅有規模約為其兩倍的模型能超越它。所有模型權重已於今日在 Hugging Face 上公開。核心資訊:規模:2.4B 參數(2.2B Qwen3.5-2B VLM + 0.2B 流匹配動作專家)。記憶涵蓋最近 3 幀,間隔 1 秒。運行環境:在 1 張 NVIDIA H100 上進行基準測試。從兩個 256×256 攝像頭輸入,編譯後每個動作區塊耗時 21.8 毫秒。權重以 BF16 格式提供,暫無量化版本。性能:在 LIBERO-Plus 和 RoboTwin Clean2Rand 基準測試中,為公開數據預訓練的 VLA 模型中最優;在 LIBERO-Plus 上達到 90.4% 的最高分。
DeepSeek 發布 V4.1-Flash-1017,DeepSWE 得分 77.0,並開放搜尋 API
DeepSeek 剛發布 DeepSeek-V4.1-Flash-1017,這是其 V4.1-Flash 模型的新檢查點,現已在 DeepSeek 網站、應用程式及 API 上線。DeepSeek 宣稱該版本在性能、任務完成速度與輸出簡潔度上均優於 9 月發布的 V4.1-Flash,並稱之為性能與成本的帕累託改進。同時,DeepSeek 也向所有開發者開放了搜尋與擷取 API。對代理建構者而言,這意味著更便宜的編碼模型加上單一 API 金鑰即可使用的第一方網路層。重點摘要:骨架參數 552B,另加 196B 的 Engram 參數;預填充階段每 token 啟用 8B 參數,解碼階段啟用 16B 參數;上下文長度達 1M tokens,最大輸出 384K tokens。運行於 DeepSeek API 的 deepseek-flash 模型。9 月版的 V4.1-Flash 權重(763B 參數,FP8)採 MIT 授權。
2026年20款Jev替代方案:挑戰TypeSafe的決策模型
TypeSafe於2026年9月15日推出Jev,這是一款決策模型而非聊天機器人。使用者可定義允許的答案,Jev將回傳標籤、分數或是否的機率。約4週內,至少20款Jev替代方案相繼問世,來自Cloudflare、OpenAI、Perplexity、Liquid AI、AWS的Strands Labs及獨立開發者。本指南比較各方案在規模、授權、硬體、價格、Jev API相容性及已發布基準測試的表現。重點摘要:規模範圍從287M(GLiNER2.5-multi-Decide)到35B-A3B(Mapika Decider),Jev規模未公開。執行環境:Von和Laya等編碼器可在CPU上運行,4B適配器適用24 GB GPU或32 GB Mac。效能:多數開源模型在困難推理上落後Jev,但在狹窄分類任務上數款與之匹敵或超越。最佳表現:Cloudflare Clef在BANKING7取得94.20分。
長鑫科技市值兩月內跌去1萬億;阿里加碼AI硬件品牌,天貓、速賣通聯動;國行版蘋果AI有望12月上線,百度、阿里參與合作丨雷峰早報
要聞提示1.長鑫科技市值兩月內跌去1萬億2.年內已至少18次故障!DeepSeek發放宕機補償,有人拿到3000元3.阿里加碼AI硬件品牌,天貓、速賣通聯動4.國行版蘋果AI有望12月上線,百度、阿里參與合作5.可靈AI新一代模型Kling 4.