{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:J2QAIDUV7JSXVG22VIX4DVT76T","short_pith_number":"pith:J2QAIDUV","canonical_record":{"source":{"id":"2605.00528","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-05-01T09:05:28Z","cross_cats_sorted":["cs.AI","cs.LG","cs.OS"],"title_canon_sha256":"29c1169a251f1f9b1d2f4ce3ac03649d8a0ad6b41c66274100ccb236ea15aa1b","abstract_canon_sha256":"4ee36fde351ce84edcf098394946a2a402b13395be3b520a29cca5a99fa7ae6b"},"schema_version":"1.0"},"canonical_sha256":"4ea0040e95fa657a9b5aaa2fc1d67ff4c6b0255c65fad3a8c048cc30ad808bbf","source":{"kind":"arxiv","id":"2605.00528","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.00528","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"arxiv_version","alias_value":"2605.00528v2","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.00528","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"pith_short_12","alias_value":"J2QAIDUV7JSX","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"pith_short_16","alias_value":"J2QAIDUV7JSXVG22","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"pith_short_8","alias_value":"J2QAIDUV","created_at":"2026-06-23T01:12:07Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:J2QAIDUV7JSXVG22VIX4DVT76T","target":"record","payload":{"canonical_record":{"source":{"id":"2605.00528","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-05-01T09:05:28Z","cross_cats_sorted":["cs.AI","cs.LG","cs.OS"],"title_canon_sha256":"29c1169a251f1f9b1d2f4ce3ac03649d8a0ad6b41c66274100ccb236ea15aa1b","abstract_canon_sha256":"4ee36fde351ce84edcf098394946a2a402b13395be3b520a29cca5a99fa7ae6b"},"schema_version":"1.0"},"canonical_sha256":"4ea0040e95fa657a9b5aaa2fc1d67ff4c6b0255c65fad3a8c048cc30ad808bbf","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T01:12:07.813739Z","signature_b64":"pg5T0ojyl3JpZN9XM4/2S3Z86+MNsf2Qld51c7A8WBP+hre9QqHQAL18MHFsesGYmhIRqTsNxxp4FTX9E9wwAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4ea0040e95fa657a9b5aaa2fc1d67ff4c6b0255c65fad3a8c048cc30ad808bbf","last_reissued_at":"2026-06-23T01:12:07.813331Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T01:12:07.813331Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.00528","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-23T01:12:07Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ODzHQ4Pzu4brIKdt7wVzoIml1S0lSN6ZLHflL9yb5j2I+SnmMCF3MtGUkLYqP8SahIOkwVI65A1UposRKBTiCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-30T10:26:27.453745Z"},"content_sha256":"e178af1694175512fb64ab8ffd5fdcb9eda82fdafbc1abf002dc6c0fb58339b5","schema_version":"1.0","event_id":"sha256:e178af1694175512fb64ab8ffd5fdcb9eda82fdafbc1abf002dc6c0fb58339b5"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:J2QAIDUV7JSXVG22VIX4DVT76T","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"SAGA: Workflow-Atomic Scheduling for AI Agent Inference on GPU Clusters","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Scheduling AI agent workflows as single units reduces task completion time by 1.64x on GPU clusters","cross_cats":["cs.AI","cs.LG","cs.OS"],"primary_cat":"cs.DC","authors_text":"Dongxin Guo, Jikun Wu, Siu Ming Yiu","submitted_at":"2026-05-01T09:05:28Z","abstract_excerpt":"AI agents execute tens to hundreds of chained LLM calls per task, yet GPU schedulers treat each call as independent, discarding gigabytes of intermediate state between steps and inflating end-to-end latency by 3-8x. We argue that this request-level abstraction is fundamentally mismatched to compound AI workloads, and propose a shift to program-level scheduling: treating the entire agent workflow (not individual inference calls) as the first-class schedulable unit. We present SAGA, a distributed scheduler that implements this abstraction through three mechanisms: (1) Agent Execution Graphs that"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On a 64-GPU cluster serving SWE-bench coding agents and WebArena browser tasks, SAGA reduces task completion time by 1.64x (geometric mean, p < 0.001) over vLLM v0.15.1 with prefix caching and affinity routing, while improving GPU memory utilization by 1.22x and achieving 99.2% SLO attainment under multi-tenant interference.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That real-world agent workflows possess enough predictable structure for Agent Execution Graphs to accurately forecast KV cache reuse across tool-call boundaries, and that the dominant use case is latency-sensitive interactive work where a 30% throughput reduction is acceptable.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SAGA reduces AI agent task completion time by 1.64x on 64-GPU clusters by scheduling at the full workflow level with execution graphs, affinity batching, and completion-time fairness.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Scheduling AI agent workflows as single units reduces task completion time by 1.64x on GPU clusters","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ec3b445e634b92610685497691f42e4ec13f6864cb746bc338fbc38482bf8857"},"source":{"id":"2605.00528","kind":"arxiv","version":2},"verdict":{"id":"7bdc6a0f-cc22-40d5-9f31-deb5a5f8e614","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T19:10:49.517777Z","strongest_claim":"On a 64-GPU cluster serving SWE-bench coding agents and WebArena browser tasks, SAGA reduces task completion time by 1.64x (geometric mean, p < 0.001) over vLLM v0.15.1 with prefix caching and affinity routing, while improving GPU memory utilization by 1.22x and achieving 99.2% SLO attainment under multi-tenant interference.","one_line_summary":"SAGA reduces AI agent task completion time by 1.64x on 64-GPU clusters by scheduling at the full workflow level with execution graphs, affinity batching, and completion-time fairness.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That real-world agent workflows possess enough predictable structure for Agent Execution Graphs to accurately forecast KV cache reuse across tool-call boundaries, and that the dominant use case is latency-sensitive interactive work where a 30% throughput reduction is acceptable.","pith_extraction_headline":"Scheduling AI agent workflows as single units reduces task completion time by 1.64x on GPU clusters"},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.00528/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T19:40:25.577467Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T18:04:38.476505Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"a0717447cd2f49e6112c5df2c7b0134046db947fb0bd82dde60e067808b4e2ab"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"7bdc6a0f-cc22-40d5-9f31-deb5a5f8e614"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-23T01:12:07Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gPNGbrC+Zt1MYTjT6TpZ3cB8Q9EVtTNZKspjlKizMDER6IpN/cduaeWthTDS9sEyfWjIAETdu+yff1F4VBmWAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-30T10:26:27.454282Z"},"content_sha256":"f985c4f1a414cd21eb63d3af936d859bf6e80fc94490056f3a0a7b0d0690d22a","schema_version":"1.0","event_id":"sha256:f985c4f1a414cd21eb63d3af936d859bf6e80fc94490056f3a0a7b0d0690d22a"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/J2QAIDUV7JSXVG22VIX4DVT76T/bundle.json","state_url":"https://pith.science/pith/J2QAIDUV7JSXVG22VIX4DVT76T/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/J2QAIDUV7JSXVG22VIX4DVT76T/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-30T10:26:27Z","links":{"resolver":"https://pith.science/pith/J2QAIDUV7JSXVG22VIX4DVT76T","bundle":"https://pith.science/pith/J2QAIDUV7JSXVG22VIX4DVT76T/bundle.json","state":"https://pith.science/pith/J2QAIDUV7JSXVG22VIX4DVT76T/state.json","well_known_bundle":"https://pith.science/.well-known/pith/J2QAIDUV7JSXVG22VIX4DVT76T/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:J2QAIDUV7JSXVG22VIX4DVT76T","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"4ee36fde351ce84edcf098394946a2a402b13395be3b520a29cca5a99fa7ae6b","cross_cats_sorted":["cs.AI","cs.LG","cs.OS"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-05-01T09:05:28Z","title_canon_sha256":"29c1169a251f1f9b1d2f4ce3ac03649d8a0ad6b41c66274100ccb236ea15aa1b"},"schema_version":"1.0","source":{"id":"2605.00528","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.00528","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"arxiv_version","alias_value":"2605.00528v2","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.00528","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"pith_short_12","alias_value":"J2QAIDUV7JSX","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"pith_short_16","alias_value":"J2QAIDUV7JSXVG22","created_at":"2026-06-23T01:12:07Z"},{"alias_kind":"pith_short_8","alias_value":"J2QAIDUV","created_at":"2026-06-23T01:12:07Z"}],"graph_snapshots":[{"event_id":"sha256:f985c4f1a414cd21eb63d3af936d859bf6e80fc94490056f3a0a7b0d0690d22a","target":"graph","created_at":"2026-06-23T01:12:07Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"On a 64-GPU cluster serving SWE-bench coding agents and WebArena browser tasks, SAGA reduces task completion time by 1.64x (geometric mean, p < 0.001) over vLLM v0.15.1 with prefix caching and affinity routing, while improving GPU memory utilization by 1.22x and achieving 99.2% SLO attainment under multi-tenant interference."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That real-world agent workflows possess enough predictable structure for Agent Execution Graphs to accurately forecast KV cache reuse across tool-call boundaries, and that the dominant use case is latency-sensitive interactive work where a 30% throughput reduction is acceptable."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SAGA reduces AI agent task completion time by 1.64x on 64-GPU clusters by scheduling at the full workflow level with execution graphs, affinity batching, and completion-time fairness."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Scheduling AI agent workflows as single units reduces task completion time by 1.64x on GPU clusters"}],"snapshot_sha256":"ec3b445e634b92610685497691f42e4ec13f6864cb746bc338fbc38482bf8857"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T19:40:25.577467Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T18:04:38.476505Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.00528/integrity.json","findings":[],"snapshot_sha256":"a0717447cd2f49e6112c5df2c7b0134046db947fb0bd82dde60e067808b4e2ab","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"AI agents execute tens to hundreds of chained LLM calls per task, yet GPU schedulers treat each call as independent, discarding gigabytes of intermediate state between steps and inflating end-to-end latency by 3-8x. We argue that this request-level abstraction is fundamentally mismatched to compound AI workloads, and propose a shift to program-level scheduling: treating the entire agent workflow (not individual inference calls) as the first-class schedulable unit. We present SAGA, a distributed scheduler that implements this abstraction through three mechanisms: (1) Agent Execution Graphs that","authors_text":"Dongxin Guo, Jikun Wu, Siu Ming Yiu","cross_cats":["cs.AI","cs.LG","cs.OS"],"headline":"Scheduling AI agent workflows as single units reduces task completion time by 1.64x on GPU clusters","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-05-01T09:05:28Z","title":"SAGA: Workflow-Atomic Scheduling for AI Agent Inference on GPU Clusters"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.00528","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-09T19:10:49.517777Z","id":"7bdc6a0f-cc22-40d5-9f31-deb5a5f8e614","model_set":{"reader":"grok-4.3"},"one_line_summary":"SAGA reduces AI agent task completion time by 1.64x on 64-GPU clusters by scheduling at the full workflow level with execution graphs, affinity batching, and completion-time fairness.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Scheduling AI agent workflows as single units reduces task completion time by 1.64x on GPU clusters","strongest_claim":"On a 64-GPU cluster serving SWE-bench coding agents and WebArena browser tasks, SAGA reduces task completion time by 1.64x (geometric mean, p < 0.001) over vLLM v0.15.1 with prefix caching and affinity routing, while improving GPU memory utilization by 1.22x and achieving 99.2% SLO attainment under multi-tenant interference.","weakest_assumption":"That real-world agent workflows possess enough predictable structure for Agent Execution Graphs to accurately forecast KV cache reuse across tool-call boundaries, and that the dominant use case is latency-sensitive interactive work where a 30% throughput reduction is acceptable."}},"verdict_id":"7bdc6a0f-cc22-40d5-9f31-deb5a5f8e614"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:e178af1694175512fb64ab8ffd5fdcb9eda82fdafbc1abf002dc6c0fb58339b5","target":"record","created_at":"2026-06-23T01:12:07Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"4ee36fde351ce84edcf098394946a2a402b13395be3b520a29cca5a99fa7ae6b","cross_cats_sorted":["cs.AI","cs.LG","cs.OS"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-05-01T09:05:28Z","title_canon_sha256":"29c1169a251f1f9b1d2f4ce3ac03649d8a0ad6b41c66274100ccb236ea15aa1b"},"schema_version":"1.0","source":{"id":"2605.00528","kind":"arxiv","version":2}},"canonical_sha256":"4ea0040e95fa657a9b5aaa2fc1d67ff4c6b0255c65fad3a8c048cc30ad808bbf","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"4ea0040e95fa657a9b5aaa2fc1d67ff4c6b0255c65fad3a8c048cc30ad808bbf","first_computed_at":"2026-06-23T01:12:07.813331Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-23T01:12:07.813331Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"pg5T0ojyl3JpZN9XM4/2S3Z86+MNsf2Qld51c7A8WBP+hre9QqHQAL18MHFsesGYmhIRqTsNxxp4FTX9E9wwAw==","signature_status":"signed_v1","signed_at":"2026-06-23T01:12:07.813739Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.00528","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:e178af1694175512fb64ab8ffd5fdcb9eda82fdafbc1abf002dc6c0fb58339b5","sha256:f985c4f1a414cd21eb63d3af936d859bf6e80fc94490056f3a0a7b0d0690d22a"],"state_sha256":"9349dcea5afd8ac29a80a90fdb4f75dec5e5061e65ba6e4623633f5161df2578"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EA3FLw1A+xEGFSMH2egNad2V/wyy6PMBY1v9qlP06O1yV65H8h36dffCiF8WAO2xIM9T8ssK/pGm3LHmNreDCQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-30T10:26:27.457724Z","bundle_sha256":"050e56d65a440aec8509985a1d01a1b5b50dff2b408a5292fd11a1dc47a07ded"}}