paperarXivTrust 82 · PrimaryPublished 1mo agoLive · 1mo ago
TraceLab: Characterizing Coding Agent Workloads for LLM Serving
Coding agents are rapidly becoming a major application of agentic LLMs, but serving them efficiently remains challenging. Progress on this challenge requires understanding real workload patterns, yet the data needed for such analysis is largely absent. Existing public traces and benchmarks do not capture real, day-to-day coding-agent usage across multiple agents and model families for serving-system analysis. To help fill this gap, we collect and release a trace of roughly 4,300 coding-agent sessions, containing about 350,000 LLM steps and 430,000 tool calls from our own day-to-day use of Clau
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- LinkedLinked via unknownagent-tools →
- LinkedLinked via unknownAgentTrace →
- LinkedLinked via unknownAgentic Resource Discovery: Let agents search →
- LinkedLinked via unknownDebugging production agents with Amazon Bedrock AgentCore Observability →
- LinkedLinked via unknownWhat's one local AI workflow you wish you'd discovered sooner? →
- FuzzySimilar title/name (fuzzy) · 59%AgentCore-8B →
“Fuzzy title match (0.73): “TraceLab: Characterizing Coding Agent Workloads for LLM Serv” ≈ “AgentCore-8B””
- PossiblePossibly related (embedding) · 28%headroomlabs-ai/headroom →
“Possibly related via embedding similarity 0.55 (not asserted). Timestamp check: artifact after paper (+3d).”
- LinkedLinked via unknownScarfBench: Benchmarking AI Agents for Enterprise Java Framework Migration →
Implements
Related to
Covers
Has model
Covers (incoming)
newsScarfBench: Benchmarking AI Agents for Enterprise Java Framework MigrationnewsREAP: Automatic Curation of Coding Agent Benchmarks from Interactive Production Usage [R]newsStructured memory filtering with metadata in AgentCore MemorynewsSentryCode: Real-time Auditor + Honeytokens for AI Coding Agents [P]newsI spent ~4.5 months building a free, self-hosted AI gateway: one endpoint for 237 providers (90+ free), auto-fallback, and a token-compression pipeline (MIT)newsI benchmarked 13 models at 65K-128K context to find out what actually matters for agentic workloadsnewsTRACE: open-source hierarchical memory for LLM agents, 82.5% on MemoryAgentBench’s EventQA using gpt-oss-20B [P]
Implements (incoming)
reposipyourdrink-ltd/bernsteinrepoautomagik-dev/genierepoagentforce314/clawcodexrepoJKHeadley/instarrepoBerriAI/litellmrepoactiveloopai/hivemindrepomm7894215/TokenTrackerrepogeneralaction/emdashrepoSakshxm1/hermes-agency-orchestratorrepobytechefhq/bytechefrepocomet-ml/opikrepoGreen-PT/honey-for-devsrepothedotmack/claude-memrepoGitlawb/zerorepoJingbiaoMei/Tokdashrepolotus-data/lotusrepogolobokov.misha/llm-review-agentsreposquirrelscan/squirrelscanrepomjason/longrepoGiskard-AI/giskard-ossrepojuyterman1000/entrolyrepotarunlnmiit/autopilot-jobhuntrepoliaohch3/claude-taprepoLazyAGI/LazyLLMrepoprasenjeet-symon/ogcoderepobosun-ai/swiftiderepoantoinezambelli/forgerepocaudena/beam_weaverrepomicrosoft/Sicoreporuvnet/agenticowrepoheng1234/claude-webrepotokentopapp/tokentoprepoSayyedhash888/memory-janitor-agentrepocall518/LogSentinelAIrepozjunlp/DataMindrepolmu-tel/peerreview-analysis-pipeline/llm-review-coderreponcz-os/tokenomicsrepoharvard-cns/orlarepoGalaxyXieyu/Awesome-Langgraph-LearnrepoLuD1161/agentjailrepoThreatRecall/zettelforgerepohollyweird/pi-deletion_scheduled-84330243repodeer-flow/llm-spacerepoFast-Editor/Lynkrrepogrisuno/ReadMenatorreponomograph/jigrepoSuppieRK/ccprepoOpenDataBox/Workspace-Benchrepozhayujie/CowAgentrepotensorflow/servingrepoNousResearch/hermes-agentrepo2FastLabs/agent-squadrepoiflytek/astron-agentrepoSWE-agent/SWE-agentrepobojieli/ai-agent-book
Related across the graph
repoSakshxm1/hermes-agency-orchestratorrepoFast-Editor/LynkrrepoLazyAGI/LazyLLMrepoGiskard-AI/giskard-ossrepoJKHeadley/instarrepoSWE-agent/SWE-agentrepojuyterman1000/entrolyrepotarunlnmiit/autopilot-jobhuntrepocaudena/beam_weaverrepogeneralaction/emdashreposipyourdrink-ltd/bernsteinrepoThreatRecall/zettelforgereporuvnet/agenticowrepoactiveloopai/hivemindrepotensorflow/servingnewsSentryCode: Real-time Auditor + Honeytokens for AI Coding Agents [P]repoGitlawb/zeronewsI benchmarked 13 models at 65K-128K context to find out what actually matters for agentic workloadsrepomjason/longreposquirrelscan/squirrelscanreponcz-os/tokenomicsrepoharvard-cns/orlarepotokentopapp/tokentopnewsStructured memory filtering with metadata in AgentCore MemorynewsDebugging production agents with Amazon Bedrock AgentCore Observabilityrepozhayujie/CowAgentrepoLuD1161/agentjailrepoNousResearch/hermes-agentrepomicrosoft/Sicorepocall518/LogSentinelAIrepomm7894215/TokenTrackerrepoJingbiaoMei/Tokdashrepoliaohch3/claude-tapnewsAgentic Resource Discovery: Let agents searchrepocomet-ml/opikrepobojieli/ai-agent-bookrepothedotmack/claude-memreponomograph/jignewsScarfBench: Benchmarking AI Agents for Enterprise Java Framework MigrationrepoGreen-PT/honey-for-devsrepoSayyedhash888/memory-janitor-agentmodelAgentCore-8BnewsTRACE: open-source hierarchical memory for LLM agents, 82.5% on MemoryAgentBench’s EventQA using gpt-oss-20B [P]repoheadroomlabs-ai/headroomnewsI spent ~4.5 months building a free, self-hosted AI gateway: one endpoint for 237 providers (90+ free), auto-fallback, and a token-compression pipeline (MIT)repolotus-data/lotusrepogrisuno/ReadMenatorrepodeer-flow/llm-spacenewsWhat's one local AI workflow you wish you'd discovered sooner?repoGalaxyXieyu/Awesome-Langgraph-Learnrepozjunlp/DataMindrepobosun-ai/swiftiderepoantoinezambelli/forgerepobytechefhq/bytechefrepo2FastLabs/agent-squadrepogolobokov.misha/llm-review-agentsrepoagentforce314/clawcodexrepoSuppieRK/ccprepolmu-tel/peerreview-analysis-pipeline/llm-review-coderrepoiflytek/astron-agentrepohollyweird/pi-deletion_scheduled-84330243repoautomagik-dev/genierepoOpenDataBox/Workspace-BenchrepoBerriAI/litellmrepoheng1234/claude-webrepoagent-toolsrepoprasenjeet-symon/ogcodenewsREAP: Automatic Curation of Coding Agent Benchmarks from Interactive Production Usage [R]toolAgentTrace