Reviews Directory

Type: kb/types/index.md

← Parent

  • ACE (agent-memory-system-review) - Three-role generator-reflector-curator loop that grows a sectioned playbook of ID-tagged bullets with helpful/harmful counters; curator prompt now only solicits ADD, with optional embedding-based merge
  • Agent Skills for Context Engineering (agent-memory-system-review) - Skill-based context engineering framework — 14 instructional modules covering attention mechanics, multi-agent patterns, memory, evaluation. Strong on operational patterns, weaker on learning theory.
  • Agent Workflow Memory (agent-memory-system-review) - Workflow-memory system for web agents that distils successful trajectories into reusable prompt workflows for WebArena and Mind2Web
  • Agent-R (agent-memory-system-review) - Self-training agent that mines MCTS action-observation trees into path-paired revision conversations, then hands the resulting JSONL off to an external fine-tuner rather than keeping a persistent memory artifact
  • Agent-S (agent-memory-system-review) - Agent-S review: versioned computer-use agents with S1/S2 JSON experience memory, S2.5/S3 prompt-time reflection, code-agent delegation, and BBoN trajectory judging
  • AgentFly (agent-memory-system-review) - Planner-executor research agent that turns benchmark answer traces into JSONL case memory and a trained case selector for planner prompt reuse
  • Agentic Harness Engineering (agent-memory-system-review) - Code-inspected review of AHE, an observability-driven coding-agent harness optimizer that mines benchmark rollouts into prompt, tool, middleware, skill, memory, and sub-agent changes
  • Amazon Science SAGE (agent-memory-system-review) - Amazon Science SAGE review: AppWorld skill-library rollouts, executable function reuse, skill-integrated rewards, and GRPO training
  • Archie (agent-memory-system-review) - Arch Linux config repo with Stow-managed multi-root deployment, Incus dev VMs, and agent-executable work-item docs; strong operational packaging, no real knowledge-learning loop
  • AriGraph (agent-memory-system-review) - AriGraph review: TextWorld agent memory that extracts observation traces into an in-run triplet graph plus episodic observations for planning, navigation, exploration, and QA retrieval
  • ARIS (agent-memory-system-review) - ARIS research-skill harness with markdown workflows, cross-model review loops, a small research wiki, verifier-backed paper audits, and trace-derived meta-optimization of its own skills
  • Ars Contexta (agent-memory-system-review) - Claude Code plugin that generates knowledge systems from conversation, backed by 249 research claims. Ancestor of our KB — we borrowed link semantics, propositional titles, and three-space architecture, then diverged in theory and structure.
  • Atomic (agent-memory-system-review) - Database-backed personal KB that stores markdown atoms in SQLite/Postgres, enriches them with embeddings/tags/semantic edges, and builds per-tag wiki plus search/chat surfaces
  • auto-harness (agent-memory-system-review) - Minimal agent-optimization harness for improving one file against tau-bench with regression-suite and held-out-score gates, keeping learnings as workshop state rather than structured evaluators
  • Autocontext (agent-memory-system-review) - Multi-role iterative control plane that turns repeated scenario runs into playbooks, typed lessons, harness mutations, and optionally distilled MLX models — now reorganized around a canonical concept model (Scenario/Task/Mission/Campaign) with a data-driven role DAG and trend-aware gating
  • Awesome Agent Memory (agent-memory-system-review) - Curated Awesome-list for agent memory papers, products, benchmarks, and surveys; useful as a discovery map but not an implemented memory system
  • Binder (agent-memory-system-review) - Local-first typed knowledge graph with markdown/YAML projections, schema-as-data, and immutable transactions; clearest reviewed example of database-first structure surfaced as editable files
  • browzy.ai (agent-memory-system-review) - Terminal TypeScript PKB that compiles raw sources into a markdown wiki, uses SQLite FTS5 as a derived retrieval index, and writes thin session-derived digests and crystallized insight drafts
  • ByteRover CLI (agent-memory-system-review) - Source-available coding-agent CLI with file-backed .brv/context-tree, tiered search/query, live scoring+archive/manifest layers, git-like context-tree VC, and multi-agent connector packaging
  • cass-memory (agent-memory-system-review) - Three-layer procedural memory (episodic/working/procedural) for AI coding agents with confidence-decayed YAML playbook, Jaccard conflict detection, cross-agent session mining, TOON/JSON output, and an MCP server
  • Claude Context Guard (agent-memory-system-review) - Claude Code continuity scaffold that persists session, task, decision, and comment state in markdown safeguard files and rebuilds context through prompt-defined recovery routines
  • ClawVault (agent-memory-system-review) - TypeScript vault memory (now v3.5, deprecated in favor of OpenClaw native memory) with write-time fact extraction, a typed observation ledger, scored promotion, hybrid search, and OpenClaw memory-slot plugin
  • Closure-SDK (agent-memory-system-review) - Geometric S3 memory and integrity runtime where ordered carrier streams promote into quaternion genome state, not readable notes or model weights
  • Cludebot (agent-memory-system-review) - Clude/cludebot memory SDK — now a pnpm monorepo with Supabase/hosted/local-JSON modes, five-type decay, six-operation dream cycle plus optional JEPA Phase 4.5, clinamen, and 9-tool MCP with mode-specific capabilities including skill extraction and confidence-gated recall
  • CocoIndex (agent-memory-system-review) - Rust-backed incremental indexing framework that compiles Python-declared dataflows into continuously maintained vector, graph, and relational targets via Postgres tracking tables
  • Cognee (agent-memory-system-review) - Pipeline-first knowledge engine (add/cognify/memify/search) with Pydantic-schema graph extraction, poly-store backends, and multi-tenancy — the strongest database-side counterexample to files-first architecture, but treats knowledge as a data engineering problem rather than a curation problem
  • Context Constitution (agent-memory-system-review) - Instruction-first governance corpus for Letta agents treating context management as identity, memory, and continuity policy; a related system defined mainly by doctrine rather than code
  • CORAL (agent-memory-system-review) - Multi-agent coding harness with git worktrees, checkpointed shared notes/skills, heartbeat prompts, and eval-gated iteration; strongest lightweight open-source outer loop for collaborative code search
  • cq (agent-memory-system-review) - Local-first agent knowledge commons with SQLite local/team stores, approval-gated team sharing, and a plugin-packaged query/propose/confirm loop; strongest reviewed shared-learning reference so far
  • CrewAI Memory (agent-memory-system-review) - CrewAI's in-framework memory layer with LLM-scoped vector records, composite recall, async save barriers, agent memory tools, and HITL lesson distillation
  • Decapod (agent-memory-system-review) - Rust governance kernel for AI coding agents that forces intent codification, proof-gated completion, and workspace isolation before code touches a repo — strongest reference for hard-oracle verification in agent workflows, but constitution documents claim transformations the code does not perform
  • DocMason (agent-memory-system-review) - Repo-native document-analysis app with strict staging/published KB boundaries, multimodal evidence channels, typed interaction memories, and a governed canonical-ask front door that owns answer admissibility
  • Dynamic Cheatsheet (agent-memory-system-review) - Test-time adaptive memory where the core stateful cheatsheet variants rewrite a carried-forward cheatsheet string; v2.0 adds a hybrid cumulative-plus-retrieval approach and server-side code execution, but curation is still full-document rewrite with no enforced invariants
  • engraph (agent-memory-system-review) - Obsidian vault server with SQLite hybrid index, wikilink graph expansion, section-level writes, and local MCP/HTTP surfaces; strongest local-first derived index over a human note substrate
  • EQUIPA (agent-memory-system-review) - Python/stdlib multi-agent coding orchestrator with git-worktree dev-test loops, SQLite operational memory, three-tier trace-derived tuning (ForgeSmith/SIMBA/GEPA), SWE-bench harness, and cumulative-knowledge warm-start
  • Exocomp (agent-memory-system-review) - Go coding-agent harness with role-scoped tools, sandboxed execution, and file-backed bug/changelog coordination; execution controls are real, but planning and sub-agent workflows are still stubbed
  • ExpeL (agent-memory-system-review) - Staged trajectory-to-rule pipeline with ADD/EDIT/REMOVE/AGREE rule updates and eval-time prompt injection; all inspectable artifacts, no weights
  • G-Memory (agent-memory-system-review) - Multi-agent memory harness that layers within-run state-graph traces, a task-similarity graph over Chroma-stored trajectories, and a scored JSON rule list with ADD/EDIT/REMOVE/AGREE maintenance
  • GBrain (agent-memory-system-review) - Personal-brain system with pluggable PGLite/Postgres engines, compiled-truth-plus-timeline pages, deterministic CLI primitives, and fat markdown skills that route integration recipes through an external agent
  • getsentry/skills (agent-memory-system-review) - Sentry's shared skills repo with a skill-writer meta-skill that codifies the skill creation process itself — source-driven synthesis with depth gates, labeled iteration, description-as-trigger optimization, and the Agent Skills cross-tool spec
  • Gnosis (agent-memory-system-review) - Repo-local Go CLI for agent-written why-memory, with JSONL entries, disposable SQLite FTS search, and doctrine-driven session capture
  • HALO (agent-memory-system-review) - HALO trace-analysis engine for optimizing agent harnesses from OTel JSONL runs, with bounded trace tools, subagent fan-out, sandboxed code analysis, and coding-agent-mediated harness edits
  • Hindsight (agent-memory-system-review) - Production agent memory backend with LLM-based fact extraction, three biomimetic fact types with partial HNSW indexes, 4-way parallel retrieval (semantic + BM25 + link-expansion graph + temporal spreading), evidence-grounded auto-consolidation, and a reflect agent whose hierarchical tools now include directive enforcement
  • HippoRAG (agent-memory-system-review) - Graph-augmented RAG library that stores document chunks, OpenIE triples, entity/fact embeddings, and an igraph PageRank retrieval graph; strong on associative retrieval, thin on governance and agent memory lifecycle
  • Hyalo (agent-memory-system-review) - Rust CLI for Obsidian-compatible markdown vaults with single-pass scanning, ephemeral MessagePack indexes, mutation-safe link operations, and one-command Claude bootstrap
  • HyperAgents (agent-memory-system-review) - Meta's self-referential agent-evolution harness using git diff lineage, Docker replay, and benchmark-scored parent selection; useful for deploy-time learning comparisons, but not a knowledge system
  • KBLaM (agent-memory-system-review) - Research implementation that converts key-value KB records into trainable attention key/value tensors, replacing external retrieval with model-internal KB attention
  • kenhuangus/llm-wiki (agent-memory-system-review) - Executable local-first LLM Wiki pipeline with Python ingestion, LLM extraction/merge, BM25 search, monitors, UI, and a partial prompt-optimization loop
  • LACP (agent-memory-system-review) - Local agent control plane for Claude/Codex/Hermes with policy-gated execution, Claude hooks, Obsidian memory automation, and provenance receipts; strongest reviewed governance-heavy local ops stack
  • LLM Wiki (agent-memory-system-review) - Claude Code plugin and portable AGENTS protocol for topic-isolated compiled markdown wikis, packaging ingest/research/query workflows as prompt artifacts rather than executable software
  • MehmetGoekce/llm-wiki (agent-memory-system-review) - Promptware LLM Wiki bootstrap kit for Claude Code with L1/L2 memory split, Logseq/Obsidian schemas, setup script, and OpenSpec requirements
  • Memori (agent-memory-system-review) - Python/TypeScript SDK plus MCP/cloud memory layer that intercepts LLM calls, mines conversations and agent traces into facts/triples/summaries, and injects filtered recall context
  • MemoryOS (agent-memory-system-review) - MemoryOS review: hierarchical conversational memory that promotes user/assistant dialogue traces into short-term buffers, mid-term sessions, long-term profiles, and retrievable knowledge
  • MemPalace (agent-memory-system-review) - Local-first Python memory system storing verbatim conversation and project chunks as ChromaDB drawers with wing/room metadata, a SQLite temporal fact graph, optional AAAK lossy summaries, 19 MCP tools, and a write-ahead log
  • MentisDB (agent-memory-system-review) - Hash-chained semantic memory ledger with binary append-only storage, additive ranked retrieval, and a diff-based signed skill registry exposed over MCP and REST
  • Meta-Harness (agent-memory-system-review) - Stanford IRIS Lab harness-code optimizer with Claude-proposed memory/scaffold variants, benchmark-gated promotion, run logs, frontiers, and executable candidate artifacts
  • MiroShark (agent-memory-system-review) - Document-to-simulation stack with Neo4j graph extraction, cross-platform agent rounds, sliding-window compaction, and ReACT reporting; strongest nearby reference for graph-backed simulation loops
  • nao (agent-memory-system-review) - Analytics-agent framework that compiles data context into project files, exposes it through file/SQL tools, and extracts persistent user memories from chat traces
  • Napkin (agent-memory-system-review) - Obsidian-compatible markdown-vault CLI/SDK with cached BM25 search, TF-IDF folder overviews, pi-based auto-distill, native pi tools, and LongMemEval benchmarks — reference for agent-shaped retrieval UX on a plain-file substrate
  • Nuggets (agent-memory-system-review) - Pi-coupled personal memory assistant with local HRR nugget files and chat-channel scheduling - strongest reference for file-backed scratch memory, though its promotion loop is only partially wired
  • o-o (agent-memory-system-review) - Polyglot HTML/bash living-document system where each file carries its own update contract, agent dispatch, rendering, and source cache — the file is the app, with no server, database, or build step
  • OpenSage (agent-memory-system-review) - ADK-based agent framework where agents create subagents and tools at runtime, with Neo4j graph memory, Docker sandboxes, and RL training hooks — strongest reference for self-modifying agent topology
  • OpenViking (agent-memory-system-review) - Context database for AI agents exposing memory, resources, and skills as a viking:// virtual filesystem with auto-generated L0/L1/L2 tiers, hierarchical recursive retrieval, and background session-commit extraction — structurally guarantees a three-tier progressive disclosure our system enforces only by convention, but its "filesystem" is a service API rather than actual files
  • Operational Ontology Framework (agent-memory-system-review) - Filesystem-first operational memory framework with Pin/Spec/Facts/Handoff/Skills artifacts and a small Python runner that promotes task learnings into project facts
  • OS-Copilot (agent-memory-system-review) - OS-Copilot/FRIDAY review: OS task agent with self-refining code execution, vector-retrieved generated tools, and course-driven tool-library accumulation
  • Pal (agent-memory-system-review) - Agno-based personal knowledge agent with a dual memory split between routing metadata and session-derived operational learnings, plus a raw-to-wiki compilation pipeline and scheduled maintenance
  • Phantom (agent-memory-system-review) - AI co-worker substrate with file-and-vector memory split, multi-block prompt assembly, and a sandbox-deny reflection subprocess mutating identity files under deterministic invariants
  • Pi Self-Learning (agent-memory-system-review) - Pi extension (v0.5.0) that auto-extracts mistake/fix learnings into a scored, git-backed-by-default memory and injects them back into the next run
  • Playground (agent-memory-system-review) - Shell-first TribleSpace agent runtime with branch-separated cognition/archive/memory, chat-log importers, and budget-adaptive temporal memory built from an append-only event graph
  • Pratiyush/llm-wiki (agent-memory-system-review) - Multi-agent session-transcript compiler that turns Claude/Codex/Cursor/Gemini history into a redacted markdown wiki, static site, exports, MCP tools, and agent prompt workflows
  • ReasoningBank (agent-memory-system-review) - Three-step pipeline — run benchmark task, auto-evaluate, extract title/description/content memory items from success or failure; embedding retrieval over prior task queries; optional parallel-trajectory self-contrast extraction
  • Reflexion (agent-memory-system-review) - NeurIPS 2023 research scaffolding where failed task attempts become short natural-language reflections that are appended to the next prompt; per-task rolling memory tail of three, no cross-task consolidation, no weight updates
  • REM is a database-heavy episodic memory service with single-pass LLM consolidation (agent-memory-system-review) - Four-database episodic memory service (Postgres + Qdrant + Neo4j + Redis) where a Go API ingests agent episodes, a Python Celery worker LLM-clusters and compresses them into scored semantic facts, and retrieval widens via temporal graph hops
  • SAGE (Sovereign Agent Governed Experience) (agent-memory-system-review) - BFT-branded agent memory with CometBFT consensus, Ed25519 signing, application-level validators, confidence decay, and encryption — consensus is ceremony in single-node mode; real value is the validation gate pattern and domain-scoped RBAC
  • Self-Training-LLM (agent-memory-system-review) - Wikipedia self-training pipeline that turns generated QA samples and hallucination scores into SFT/DPO datasets for weight updates, with no persistent memory artifact
  • Semiont (agent-memory-system-review) - Document-grounded annotation kernel with W3C annotations, git-backed events, working-tree URIs, and shared human/agent flows; strongest example here of annotation-first KB infrastructure
  • sift-kg (agent-memory-system-review) - LLM-powered document-to-knowledge-graph pipeline with schema discovery, human-in-the-loop entity resolution, and interactive visualization
  • Siftly (agent-memory-system-review) - Next.js + SQLite bookmark ingestion system whose deterministic-first, resumable enrichment pipeline offers concrete patterns for scaling KB source loading with explicit progress state
  • SignetAI (agent-memory-system-review) - Local-first cross-harness memory daemon with SQLite/FTS/vector/graph recall, trace-derived fact extraction, transcript retention, MCP tools, and connector packaging
  • SkillNote (agent-memory-system-review) - Self-hosted skill registry with partial local-first editing resilience, dual version tracks, live MCP tool exposure, and agent-submitted per-version ratings
  • SkillWeaver (agent-memory-system-review) - SkillWeaver review: web-agent exploration loop that distills successful Playwright trajectories into reusable Python API skills and shipped SkillNet files
  • SkillX (agent-memory-system-review) - SkillX review: trajectory-derived plan, functional-skill, and atomic-skill KB construction with LLM extraction, filtering, merging, and partial inference support
  • Spacebot (agent-memory-system-review) - Rust concurrent agent framework whose process-type architecture (channels, branches, workers, cortex) is the cleanest production implementation of code-level scheduling over bounded LLM calls among reviewed systems
  • Stash (agent-memory-system-review) - Postgres/pgvector MCP memory service that turns episodes into facts, relationships, causal links, goals, failures, hypotheses, and confidence decay
  • Supermemory (agent-memory-system-review) - Monorepo that open-sources Supermemory's MCP/SDK integration layer while delegating core memory extraction, contradiction handling, and profile synthesis to hosted /v3 and /v4 APIs
  • Synapptic (agent-memory-system-review) - Beta Python CLI that mines Claude Code JSONL sessions into a weighted nine-dimension profile, runs WITH/WITHOUT guard ablations with LLM-as-judge scoring, and compiles guards into nine assistant-specific memory surfaces with per-model filtering
  • Tendril (agent-memory-system-review) - Tauri/Strands sandbox that turns live user requests into persistent Deno TypeScript capabilities in a workspace-local registry
  • Thalo (agent-memory-system-review) - Custom plain-text language for knowledge management with Tree-Sitter grammar, typed entities, 27 validation rules, and LSP. Makes the same programming-theory-over-psychology bet we do, but went further into formalization with a custom DSL.
  • Tolaria (agent-memory-system-review) - Mac and Linux markdown-vault app with git-backed files, type lenses, saved views, agent context snapshots, MCP tools, and managed guidance
  • Tracecraft (agent-memory-system-review) - S3-backed CLI coordination layer for multi-agent systems — cleanest exemplar of coordination-by-convention, where coordination semantics live in naming conventions and client compliance rather than enforcement
  • Virtual Context (agent-memory-system-review) - Proxy-owned context virtualization layer that compacts conversation and tool traces into durable summaries, facts, and demand-paged topic memory instead of treating retrieval as an additive sidecar
  • Voiden (agent-memory-system-review) - Git-native API workspace whose .void files combine markdown docs, structured request blocks, linked reuse, local history, and installable agent skills
  • Voyager (agent-memory-system-review) - Embodied lifelong-learning Minecraft agent whose action/critic/curriculum/skill-manager loop promotes successful trajectories into a Chroma-indexed library of reusable JavaScript functions
  • WUPHF (agent-memory-system-review) - Local multi-agent office with fresh-session runners, git-backed team wiki, per-agent notebooks, trace-to-fact extraction, playbook synthesis, and structured team skills
  • xMemory (agent-memory-system-review) - Research-code dialogue-memory system that distills conversations into episodes, semantic facts, and LLM-summarized themes, then retrieves top-down with coverage selection plus entropy-gated episode inclusion
  • Zikkaron (agent-memory-system-review) - MCP memory server for Claude Code: 26 neuroscience-branded subsystems implemented as heuristic Python without LLM calls — vocabulary over mechanism, but compaction hooks and WRRF retrieval fusion are genuinely borrowable