Parent mission
The coordinator owns accepted project history, decisions, handoff, and Memory Checkout state.
Event-sourced memory for agent work
Zaxy keeps multi-agent projects replayable: parent missions, isolated worker sessions, cited findings, purpose-conditioned checkout, conflict review, and accepted merge-back into one durable project history.
runtimeembedded Kuzu default
sourceEventloom append-only JSONL
checkoutcited, purpose-conditioned context
Coordinate
Spawning agents is easy. The hard part is turning isolated investigations into one trustworthy state of work. Zaxy records each worker in its own Eventloom session, reviews findings with evidence, marks stale and conflicting claims, and promotes only accepted facts into the parent mission.
The coordinator owns accepted project history, decisions, handoff, and Memory Checkout state.
Agents investigate in isolated logs, so exploration does not contaminate authoritative memory.
Human or coordinator-agent review accepts, rejects, defers, or promotes findings with cited provenance.
Architecture
Missionobjective and parent state
Workersisolated Eventloom sessions
Findingsevidence, confidence, citations
Reviewconflicts, stale claims, approvals
Checkoutaccepted cited prompt state
Eventloom source of truth remains the append-only project record. The
default local runtime is embedded Kuzu, launched and cleaned by
zaxy init and zaxy doctor. Neo4j remains the
sidecar control backend; pgGraph, LatticeDB, and Pathlight are
advanced integration tracks for teams that need alternate deployment or
observability posture.
Purpose control plane
Zaxy now carries purpose through retrieval, checkout diagnostics, feedback, compaction, and Coordinate accepted state. This is still framed as project-local agent work memory, not a broad Company Brain claim.
Interfaces
coordination_checkout accepted parent state plus diagnostic worker-local findings
coordination_approval_packet reviewable accept/reject/defer/promote payloads
memory_checkout answerability, current_citation_count, required action, and memory_feedback guidance
CoordinationAdapter dependency-light Python wrapper with LangGraph and CrewAI helper paths
dashboard --enable-coordinate-review opt-in human review controls over replay-backed state; read-only remains the default
Benchmark evidence
Current artifacts separate external benchmark evidence, internal release gates, and unsupported comparative claims. Zaxy now has a full Harvey LAB external memory-ablation run across all ten pinned legal agent tasks: mean criterion pass rate 0.788, +0.184 versus regular/no-memory, +0.081 versus the article-best task rows, 9/10 task wins, and a head-to-head win over the strongest GBrain task rows. CoordinationBench v1 still passes accepted-finding precision and recall, conflict handling, stale rejection, proof-backed synthesis, and parent answerability. Larger memory-system comparisons against Mem0, Agent Memory, Zep/Graphiti, and MemPalace remain disclosure-only until pinned same-harness adapters produce locally scored, fingerprinted results.
Full ten-task external legal-agent memory-ablation run, +0.184 versus regular/no-memory and 9/10 task wins versus article-best rows.
Purpose recall, ontology shift, outcome feedback, evidence discipline, and neutral-substrate projection are covered by purpose-v1.
Accepted-state synthesis, non-authoritative leakage, citation coverage, parent answerability, and Eventloom replayability pass.
Harvey LAB is published as scoped external evidence. Larger memory-system comparisons stay disclosure-only until same-harness audits are complete.
These rows are release evidence and disclosure status, not a universal memory leaderboard.
| Artifact | Status | What it supports | What it does not support |
|---|---|---|---|
| Harvey LAB external memory-ablation | complete | Full 10-task legal-agent benchmark evidence: 0.788 mean criterion pass rate, +0.184 vs regular/no-memory, +0.081 vs article-best rows, 9/10 task wins. | Same-harness full-suite scores for non-Zaxy systems beyond the article-published matrix. |
| purpose-v1 | passed | Purpose-conditioned retrieval, checkout, feedback, and compaction contract. | External superiority claims against larger memory systems without pinned same-harness adapters. |
| public-derived purpose holdout | diagnostic | Representative pressure cases with citation coverage and frozen fingerprint. | Release gate or outside-user validation. |
| CoordinationBench real-v1 | passed | Accepted parent state, conflict handling, stale rejection, proof-backed synthesis. | Generic RAG, open-domain QA, or all memory workflows. |
| Semantic Reach / Quarq | blocked | Small-project unsupported disclosure rows retained for audit completeness. | Headline competitive positioning or completed competitor scoring. |
| BM25 worker logs | 0.666667 precision | Same-harness local baseline over rendered worker findings. | Accepted project state, conflict recall, stale rejection, or citation coverage. |
| Markdown notes | 0.636364 precision | Same-harness local baseline without promotion or conflict semantics. | Authoritative parent checkout or proof-backed merge-back. |
| Flat transcript | 0.272727 precision | Contamination baseline that treats all worker output as accepted state. | Reliable accepted-state precision, stale rejection, or conflict review. |
| Mem0 / Agent Memory / Zep-Graphiti / MemPalace / GBrain | not_run | Primary external comparison set: memory-system disclosures plus Harvey LAB GBrain task rows. | Same-harness memory-system scores until pinned runner manifests exist. |
Install
pipx install zaxy-memory
zaxy init
zaxy memory log --eventloom-path .eventloom --limit 5
zaxy memory bootstrap --eventloom-path .eventloom
zaxy doctor --eventloom-path .eventloomzaxy coordinate start "ship auth refactor" --mission auth-main
zaxy coordinate worker create --mission auth-main --worker auth-api
zaxy coordinate assign --mission auth-main --worker auth-api "trace failures"
zaxy coordinate brief --mission auth-main
zaxy coordinate checkout --mission auth-mainZaxy writes `.env.local`, records session genesis and heartbeat, checks graph posture, and prints the MCP command or config path.
Session history lives in .eventloom/ as append-only JSONL. The graph is a rebuildable projection.
memory log, memory bootstrap, doctor, and hook-status expose Last checkout, capture, and stale-memory posture.
Documentation
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。