| LACP |
LACP combines a native agent loop, host launch controls and trace-derived memory, with path-specific enforcement and unobserved improvement benefits. |
e31f134b689afd6a71d94b0420682d1f528d6fa1 |
AAS-2026-10-14-lacp-91efd16f7a35-01 |
2026-10-14 |
code-grounded |
| AgeMem |
AgeMem trains one policy to edit stored memory and active context; its paper specifies later consumption and reports bundled gains, without inspectable execution or preservation guarantees. |
arxiv-2601.01885v1-html-2026-10-12 |
AAS-2026-10-12-2601-01885v1-32274e5bddc1-01 |
2026-10-12 |
doc-grounded |
| trajectory-informed-memory-generation |
Trajectory-informed memory generation turns completed agent experience into consolidated prompt guidance; reported AppWorld gains are configuration-specific, without verified causal attribution or an exercised improvement cycle. |
arxiv-2603.10600v1-html-captured-2026-10-14 |
AAS-2026-10-14-2603-10600v1-a956324c9a82-01 |
2026-10-14 |
doc-grounded |
| wikiskill |
WikiSkill retains trace-derived wiki explanations and rejected interventions to guide validation-selected skills; its paper reports gains but does not expose executable enforcement or full-loop traces. |
arxiv-2608.27454v1-text-2026-10-15 |
AAS-2026-10-15-2608-27454-65116a9d0de7-01 |
2026-10-15 |
doc-grounded |
| aide2 |
AIDE2 selects model-written research-agent harnesses by private task grades, with reported whole-harness gains but unresolved component effects and self-improver acceleration. |
arxiv-2609.26457v1-2026-10-14 |
AAS-2026-10-14-2609-26457-6b3bcd4791cf-01 |
2026-10-14 |
doc-grounded |
| origin |
Origin retains and reorganizes local knowledge, checks page support and serves later context; activation, semantic accuracy and improved future action remain unobserved. |
61b462ea6adf019b3e27d3f58ac5a0fe055f76b1 |
AAS-2026-10-15-origin-805d03bc599f-01 |
2026-10-15 |
code-grounded |
| MiroShark |
MiroShark turns graph-derived personas and simulated activity into evolving agent context and reports; source inspection supports adaptation, not calibrated forecasts or demonstrated improvement. |
e95d3d8c2a2206b8b6695648e8504cf23436d3de |
AAS-2026-10-15-miroshark-b3df0e053a29-01 |
2026-10-15 |
code-grounded |
| Cortex |
Cortex retains typed knowledge for requested retrieval, optional model enrichment and fixed-rule inference; access-derived tiers do not establish better ranking or task performance. |
a51c2209eb6d26812c19c97aaade3cb64ccfc23c |
AAS-2026-10-17-cortex-3da14ce55a2b-01 |
2026-10-17 |
code-grounded |
| RSIAgent |
RSIAgent connects model-driven task verification and curriculum practice to a reusable experience bank; publication is wired, but semantic uptake and causal improvement are unobserved. |
4d7ef43770dda341fccd9a5743db92407687c615 |
AAS-2026-10-14-rsiagent-b11225763c0b-01 |
2026-10-14 |
code-grounded |
| llm-wiki-coordination |
LLM Wiki Coordination preserves peer-reviewed dialogue and revision-specific canon endorsements through agent/user protocols; its read-only audit diagnoses selected structural faults, not semantic truth. |
126749634d1c8c2fd6141f59711c882c2d629699 |
AAS-2026-10-14-llm-wiki-coordination-a12f3c7007f2-01 |
2026-10-14 |
code-grounded |
| Agentic Local Brain |
Agentic Local Brain wires collected documents, mined state, wiki prose and conversation history into later retrieval and browsing, without demonstrated factual acceptance or improvement. |
41ea0cb104686eb0d39f32c94c2b84d37cdff0e7 |
AAS-2026-10-12-agentic-local-brain-f00cbd577178-01 |
2026-10-12 |
code-grounded |
| AgentFly |
AgentFly couples serial planner–executor tool work with outcome-labelled case reuse and separate retriever training; wiring supports future guidance, not demonstrated improvement. |
42fbbcac63dd58ed6856c0761357345a58e4f032 |
AAS-2026-10-12-agentfly-bbfc9fb2d5c7-01 |
2026-10-12 |
code-grounded |
| Ars Contexta |
Ars Contexta derives and operates a persistent Markdown knowledge vault through model-facing procedures and shell hooks; integration gaps and missing runs limit completion and improvement claims. |
2acfd5cc4473c4d06c46be63df748e77e00e2746 |
AAS-2026-10-12-arscontexta-f9e5f2f34166-01 |
2026-10-12 |
code-grounded |
| Agno |
Agno owns model/tool execution and orchestrated handoffs, with plural retained-context routes and scoped admission checks; successful work and improvement remain unobserved. |
00497422d4a386d93ab3b45c1a11faf42fcc932f |
AAS-2026-10-13-agno-60e6e09dd4a6-01 |
2026-10-13 |
code-grounded |
| pal |
Pal coordinates source ingestion, wiki compilation and later retrieval, with configured experience-to-guidance updates but no observed fidelity, reuse or improvement. |
6516b8ede0c085e48f39f3bd04cb85b475a855dc |
AAS-2026-10-17-pal-e313a0902289-01 |
2026-10-17 |
code-grounded |
| claude-obsidian |
claude-obsidian combines host-directed knowledge workflows with journaled vault writes, provenance gates and bounded read-back; semantic correctness and compounding benefit remain unobserved. |
32ac5a02c4e082e4a5628ca810776375e134708e |
AAS-2026-10-13-claude-obsidian-c5e56e25c8bb-01 |
2026-10-13 |
code-grounded |
| SkillRL |
SkillRL wires failure-fed textual skill admission and reward-driven policy fitting into later rollouts, without establishing skill effectiveness or causal improvement. |
4d984d455b9b02f0bad283cbe9ab3d66a7e5b89d |
AAS-2026-10-14-skillrl-0dba0e1c895e-01 |
2026-10-14 |
code-grounded |
| AriGraph |
AriGraph turns game observations and QA paragraphs into retrieved facts and episodes; model-guided withdrawal changes live facts without validating truth or removing historical text. |
e884b76d7fa5185a3a8a55e5a67393b5a43f5ef2 |
AAS-2026-10-12-arigraph-fd0f6f1c1383-01 |
2026-10-12 |
code-grounded |
| llm-project-wiki |
LLM Project Wiki specifies model-authored project pages, wiki-first lookup and gap-driven repair; its prompt package does not demonstrate deployed maintenance or improved answers. |
1fdfba9129a46afd18aa68a83f12d6716ab694c6 |
AAS-2026-10-14-llm-project-wiki-c6c664168f3e-01 |
2026-10-14 |
doc-grounded |
| Stash |
Stash retains caller reports, curates them through model-driven maintenance and serves selective memory; structural admission does not guarantee truth or effective withdrawal. |
d34ed430c4348ef741ef95146b488aea8616f4fb |
AAS-2026-10-15-stash-7ac5612837a0-01 |
2026-10-15 |
code-grounded |
| AI-Context-OS |
AI-Context-OS selects local Markdown context and feeds access traces into later ranking, but admission does not establish truth and improvement remains unobserved. |
9ed263080cadf98702bebb42d5260b45bd526f6c |
AAS-2026-10-18-ai-context-os-82777c0c3926-01 |
2026-10-18 |
code-grounded |
| echel |
Echel selects related product records for Hermes and provides canonical review and staleness controls, but those separate paths do not establish an integrated learning loop. |
aee36de4d2d1c404fb36e9ad68cc0a920784ee61 |
AAS-2026-10-15-echel-08d27848b99a-01 |
2026-10-15 |
code-grounded |
| Zikkaron |
Zikkaron turns retained text and activity traces into selected host context, with automatic memory updates but no observed capacity gain or complete replay guarantee. |
dda34a5d903d04ecb5517af214d437873c833302 |
AAS-2026-10-15-zikkaron-0102ec63c6e5-01 |
2026-10-15 |
code-grounded |
| SAGE |
SAGE couples retained Python skills with reward-fed policy updates, but its shipped two-node launch has a loader mismatch and improvement remains unobserved. |
3c9244e82244abb1adc5467ee601a03ba0f433a0 |
AAS-2026-10-15-sage-d0de15ae5730-01 |
2026-10-15 |
code-grounded |
| quicky-wiki |
Quicky Wiki automatically compiles documents into a confidence-scored graph for later answers, but factual acceptance, complete withdrawal and improvement remain unestablished. |
cfeecbffce5b6d0f16533b258a1e88edd86f731e |
AAS-2026-10-15-quicky-wiki-628f7d7cc628-01 |
2026-10-15 |
code-grounded |
| Apache Maka |
Apache Maka combines Host-owned durable execution with bounded context, delegated work and evidence-gated memory; successful operation and improvement remain unobserved. |
1751c399239b0f6b717e98e2a5e2b875c3dd11ce |
AAS-2026-10-15-maka-4a8b73e41579-01 |
2026-10-15 |
code-grounded |
| Synapptic |
Synapptic turns coding-session conversations into reusable profiles and preventive instructions, with heuristic admission and synthetic feedback but no demonstrated downstream improvement. |
bcde026770daa5cd9a8ca18ca7182b0aeaa87d1f |
AAS-2026-10-15-synapptic-9dce97c66546-01 |
2026-10-15 |
code-grounded |
| Tracecraft |
Tracecraft preserves bucket-backed coordination state and host transcripts through explicit reads, with cooperative task gates rather than verified work or automatic model-context restoration. |
97c6b7447f8e696f0116357d9f86bb0ece25e5b1 |
AAS-2026-10-14-tracecraft-9c4af961e8e2-01 |
2026-10-14 |
code-grounded |
| arsumbris |
Arsumbris declares file-backed typed knowledge, graph-driven views and profile-mediated agent actions; its entry repository implements bootstrap configuration, not the downstream knowledge loop. |
716bb14f2bf50e249d414525c8e07c50478fc786 |
AAS-2026-10-15-arsumbris-a9be6bac297f-01 |
2026-10-15 |
doc-grounded |
| llm-context-base |
llm-context-base retains knowledge and conversation-derived guidance in files for selective assistant reload; its prompts prescribe adaptation, but operation and improvement are unobserved. |
6d01cba8e2c22f9ca2519c70073a05cd54378a8c |
AAS-2026-10-14-llm-context-base-f63a7c0fb5f2-01 |
2026-10-14 |
doc-grounded |
| Claude Context Guard |
Claude Context Guard requests model-maintained project ledgers and selective cross-session reload, with human-gated lessons and narrow shell backups, but no demonstrated loss-free recovery or improvement. |
efcd0a1815887483a862b32bfb3fbfd6aac79aa6 |
AAS-2026-10-13-claude-context-guard-357cd90664df-01 |
2026-10-13 |
code-grounded |
| Synthadoc |
Synthadoc compiles sources into a persistent wiki and maintains it through route-specific checks and edits; factual acceptance and successful self-improvement remain unestablished. |
b80d9388a3013e8051baf2854c7250942c46e91a |
AAS-2026-10-15-synthadoc-a3cae9152fff-01 |
2026-10-15 |
code-grounded |
| TheKnowledge |
TheKnowledge turns sources into reusable wiki context through code-owned model calls and route-specific admission, with wired feedback reuse but unobserved grounding and improvement. |
aca6e29e69c6a72b697b8178841d03e6296ee3fc |
AAS-2026-10-14-theknowledge-319a64927ee3-01 |
2026-10-14 |
code-grounded |
| MemoryOS |
MemoryOS automatically derives persistent dialogue context and reconstructs later prompts; implementation supports reuse, not verified factual acceptance or improved dialogue. |
587ed7755c7aed179965792830ff1b5ad9a6fa92 |
AAS-2026-10-16-memoryos-bb72a1b70e03-01 |
2026-10-16 |
code-grounded |
| Basic Memory |
Basic Memory retains Markdown knowledge and supplies host-specific continuation context; implemented capture and checks do not establish factual acceptance or improved future action. |
a72751c5d626380dedeed30f55f7a80ba3dc710d |
AAS-2026-10-16-basic-memory-46fe4c6cfdc3-01 |
2026-10-16 |
code-grounded |
| beever-atlas |
Beever Atlas turns conversations into searchable facts, graph data and maintained wiki pages; consumer-specific controls and unobserved outcomes limit freshness and improvement claims. |
7d791af27ef4aa19644cb385c7ca24619bc4f848 |
AAS-2026-10-15-beever-atlas-10a3512c094c-01 |
2026-10-15 |
code-grounded |
| GMemory |
GMemory retains task experience and automatically revises reusable advice for multi-agent benchmarks; its wired adaptation does not establish improved performance or validated causal rules. |
7b581c51d993bd600df14691d101d7e601040cc6 |
AAS-2026-10-15-gmemory-a2192cc823f7-01 |
2026-10-15 |
code-grounded |
| AI Agents in Depth |
AI Agents in Depth wires separate context, user-memory and retry-policy experiments; retained context and candidate checking do not establish task correctness or live self-improvement. |
dbc046eb896ac4e39aa19c7774c8bf49583b89a6 |
AAS-2026-10-12-ai-agent-book-d56e8121d636-01 |
2026-10-12 |
code-grounded |
| Agent-R |
Agent-R builds environment-scored search trees and synthetic correction examples, but export defects and missing training machinery leave iterative improvement unestablished. |
82fcc1ca7873b460949ed49022146dd988c32e31 |
AAS-2026-10-12-agent-r-216224d4270a-01 |
2026-10-12 |
code-grounded |
| ByteRover CLI |
ByteRover CLI separates caller-authored memory, structural admission and deterministic retrieval, with activity-fed ranking and continuation memory but no observed improvement. |
1052ac1a5dd0fde4da8693d4712064f7876c269c |
AAS-2026-10-15-byterover-cli-71610aaf716b-01 |
2026-10-15 |
code-grounded |
| oh-my-pi |
oh-my-pi controls coding sessions, bounded child returns and backend-specific retained guidance; trace-fed updates are wired, but operation and improved capacity are unobserved. |
2431e8a8716a4da73ae327a412f6d54c22f681e5 |
AAS-2026-10-14-oh-my-pi-c0ffc54ff768-01 |
2026-10-14 |
code-grounded |
| ATLAS |
ATLAS combines indexed logistics records with model-maintained Markdown and a chat tool loop; automatic retention and read-back are wired, but semantic acceptance and improved capacity are unestablished. |
65b91b4fd5d703e8a82ad8fa2a9b8d2f38754988 |
AAS-2026-10-12-atlas-b28ea1ea613f-01 |
2026-10-12 |
code-grounded |
| claude-workstream-kit |
Claude Workstream Kit preserves work in Git-backed files, reloads pointers and prescribes evidence checks; model compliance and improvement remain unobserved. |
15705c38a4fffb206b868b3e6bc5ffb4fd8b9d64 |
AAS-2026-10-15-claude-workstream-kit-103204afa1ab-01 |
2026-10-15 |
code-grounded |
| Tardigrade |
Tardigrade drives event-backed model/tool loops with replay and summary-based context; controls depend on composition, and improvement remains prescribed rather than demonstrated. |
1b9f1e5760d1d4d4a63eb1f87af9e5a2341de7e8 |
AAS-2026-10-15-tardigrade-455e1aaccd4d-01 |
2026-10-15 |
code-grounded |
| llmwiki-marimo |
LLM Wiki turns documents and saved conversations into reusable Markdown, with lexical retrieval, mode-specific grounding and advisory verification rather than demonstrated factual acceptance. |
0df06b0499a6242a4a5edfb1495e1d55a3e3c8b7 |
AAS-2026-10-18-llmwiki-marimo-58054ccd9ae7-01 |
2026-10-18 |
code-grounded |
| MentisDB |
MentisDB retains hash-linked memories and versioned skills for selective later context; numerical trace-fed retrieval updates are wired, but lossless continuity and improved task capacity are unestablished. |
4c9b95041da68c5a26dece9f8776ff54b05b910c |
AAS-2026-10-14-mentisdb-a8b5903ae3e2-01 |
2026-10-14 |
code-grounded |
| llm-wiki |
LLM Wiki delegates persistent Markdown compilation and revision to an external agent; its toolkit implements selective retrieval and structural checks, not semantic acceptance or demonstrated improvement. |
a424e6875aa2d5736d9c0d03d7c28bdbe32892a5 |
AAS-2026-10-14-llm-wiki-b437987c5802-01 |
2026-10-14 |
code-grounded |
| CocoIndex |
CocoIndex maintains derived indexes and reusable computation, with shipped search and typed extraction paths but no demonstrated factual acceptance or agent benefit. |
1639793b3602408a9cb220728fb30e306d433473 |
AAS-2026-10-15-cocoindex-3514227a0537-01 |
2026-10-15 |
code-grounded |
| Halo |
Halo couples model-directed trace diagnosis with retained follow-up context and user-confirmed report handoff; external harness improvement remains a partial loop. |
b7f8509745d67b499b4e80efe20ea37c03426a74 |
AAS-2026-10-14-halo-c4ed89d7f99f-01 |
2026-10-14 |
code-grounded |
| Exocomp |
Exocomp orchestrates role-specific software workers and restores project context, but report delivery and structural signoff do not certify correct work. |
b36f7ba1f3e2eee25bbcd2711db7d2d5d4824251 |
AAS-2026-10-14-exocomp-692bb27c35f9-01 |
2026-10-14 |
code-grounded |
| CrewAI |
CrewAI schedules agent work and reuses extracted context and feedback-derived instructions, but its wired checks do not establish factual warrant or demonstrated improvement. |
3da2d4e9932b5f2fa9aa73ea9245e0f8b5c154f8 |
AAS-2026-10-15-crewai-62e53db969a7-01 |
2026-10-15 |
code-grounded |
| Ecdysis |
Ecdysis turns failure signatures into collaborative repair proposals and score-selected harnesses; real editing, generalization and review-attributable improvement remain unobserved. |
cf93866d545b0974dbb0bc83b39c31fbbdeeecb8 |
AAS-2026-10-13-ecdysis-bb4c64dbdf60-01 |
2026-10-13 |
code-grounded |
| Tortoise |
Tortoise turns session traces into graph memory and supplies bounded pointers to later model work; source addressing and correction controls do not establish factual warrant or improved performance. |
54922a800f96858935b58ef250da569a4896df44 |
AAS-2026-10-30-tortoise-0c919b4ca77d-01 |
2026-10-30 |
code-grounded |
| Decapod |
Decapod supplies repository context, retained work and exact-candidate publication checks to a host agent; covered governance does not establish host compliance or improvement. |
067b8f55071e13dd103779998168b4c9ffd31a79 |
AAS-2026-10-14-decapod-c44705119841-01 |
2026-10-14 |
code-grounded |
| smriti-mcp |
Smriti MCP provides caller-led Markdown memory, lexical recall and durable access reinforcement, without owning semantic review or demonstrating agent benefit. |
8264bf9103477d9521e57b27c81dcc2e943a3230 |
AAS-2026-10-14-smriti-mcp-14f6928f9817-01 |
2026-10-14 |
code-grounded |
| antinomia |
Antinomia combines human-admitted Obsidian notes, bounded model comparison and persistent pair suppression; its checks do not establish semantic truth or improved capacity. |
6fafd01082fbecb2a18803500a1abe70a2bdc9f4 |
AAS-2026-11-08-antinomia-19694628edc2-01 |
2026-11-08 |
code-grounded |
| Engraph |
Engraph turns Markdown vaults into agent context through local retrieval, heuristic identity extraction and move-fed placement adaptation; truth, freshness and improvement remain bounded. |
f9a95bc96accc792c02ee384d9e6bf768a88c8c8 |
AAS-2026-10-14-engraph-0cd4f00695b0-01 |
2026-10-14 |
code-grounded |
| beads_rust |
beads_rust retains dependency-aware issue memory and selectively delivers work context, with conditional transition gates but no demonstrated host compliance or improvement. |
53f7a3d30aca8cab3c3e4abdefeade459e8a3089 |
AAS-2026-10-16-beads-rust-e82d5506d2ff-01 |
2026-10-16 |
code-grounded |
| cass_memory_system |
cass_memory_system turns coding sessions into retained guidance and feedback-driven standing, with conditional historical-support checks but no observed learning benefit or host compliance. |
8aca3bbb6732321dca041b32c155617784687897 |
AAS-2026-10-13-cass-memory-system-30c18219cf68-01 |
2026-10-13 |
code-grounded |
| eigenius |
Eigenius combines replayable prose formalization with separately admitted typed resources and consequential formal-proof credit, without established scientific fidelity or improvement. |
836498ebd99733067d45aa6f7b09d4ff5035dd99 |
AAS-2026-10-15-eigenius-fe68ed89519b-01 |
2026-10-15 |
code-grounded |
| LLM-WIKI-MCP |
LLM-WIKI-MCP combines local Markdown retrieval, code-directed answering and retained conversation context; its checks establish availability and structure, not factual acceptance or demonstrated improvement. |
25efb7631082fa3732723fee08e24d39b43311fa |
AAS-2026-10-15-llm-wiki-mcp-5dff0dd0576a-01 |
2026-10-15 |
code-grounded |
| Compound Engineering Plugin |
Compound Engineering Plugin combines host-interpreted coding workflows, reusable project reasoning and executable external-worker gates; compliance, useful reuse and improvement remain unobserved. |
50532b00de523980010d97fe950965b35382a3b7 |
AAS-2026-10-13-compound-engineering-plugin-6ac79f4e7555-01 |
2026-10-13 |
code-grounded |
| exo |
Exo combines durable turn execution with harness-specific context, retained procedures and conditional self-change; successful recovery and improvement remain unobserved. |
423d7d0f4394a1ddc9955d13c73c9f8a9c4d4e7e |
AAS-2026-10-15-exo-f0905e2cf44b-01 |
2026-10-15 |
code-grounded |
| Hyperagents |
Hyperagents replays model-edited agent machinery and uses benchmark archives to select successors; trace-fed guidance and policy training are wired, but exercised improvement is unobserved. |
59a68f672dfb92c74aeb7e61535d776fb36e172d |
AAS-2026-10-14-hyperagents-dd9d501b160c-01 |
2026-10-14 |
code-grounded |
| Closure-SDK |
Closure-SDK's Rust Closure EA adapts retained quaternion memory through feedback and consolidation; later reads are wired, but learning benefit and semantic warrant remain unestablished. |
9735f48a6ddd00b6f429a91443489531b3a115ef |
AAS-2026-10-13-closure-sdk-294e6865ca51-01 |
2026-10-13 |
code-grounded |
| pi-posthorse |
Pi Posthorse replaces eligible active windows with bounded continuation records and exposes notes/history recall; its checks admit context transitions, not verified task progress. |
307ef6bf05bc8bd8d718725761ffdf2f2ef72539 |
AAS-2026-10-14-pi-posthorse-58b9647d2f29-01 |
2026-10-14 |
code-grounded |
| incremental-self-improvement |
Incremental self-improvement changes its own instruction-selection policy and filters changes by historical reward rate, with reported toy-task gains but no independently inspected execution. |
fki198-94-revised-1995-01-31-captured-2026-10-14-text |
AAS-2026-10-14-fki198-94-pdf-90fb0a2a9508-01 |
2026-10-14 |
doc-grounded |
| From Agent Memory to Portable Skills |
NAMS describes trace-fed skill construction with source checks, human publication review and local grounding repair; execution, semantic warrant and improvement remain unverified. |
neo4j-article-2026-10-18-99f91d4da959 |
AAS-2026-10-18-from-agent-memory-to-portable-skills-4a509c7732b1-01 |
2026-10-18 |
doc-grounded |
| GBrain |
GBrain combines explicit memory, scoped publication, unsolicited context and feedback-driven ranking; source-grounding checks are conditional and operational benefit is unobserved. |
da434385572e0dc0d825649179aabed4107d6e60 |
AAS-2026-10-11-gbrain-9de8e3c225d3-01 |
2026-10-11 |
code-grounded |
| archie |
Archie's shell profiles update external model-role configuration before agent launch; runtime internals and the planned documentation assistant remain unestablished. |
76caf94548cf76f2399d9a7446da15e0daa22685 |
AAS-2026-10-15-archie-1ab3d8b8ff03-01 |
2026-10-15 |
code-grounded |
| JEPA-Anything |
JEPA-Anything guides a host model from world-model tasks to checked designs and code skeletons; its deterministic gates establish declared completeness, not learned performance. |
c6e6c88f3ef75a4ce7acd660d6fa5779d995512c |
AAS-2026-10-15-jepa-anything-434b01ab7774-01 |
2026-10-15 |
code-grounded |
| nao |
nao encloses SQL-backed analytics with trace-fed user guidance, lossy continuation summaries and feedback-to-context proposals; operation and improvement remain unobserved. |
0085625a8fd22e4f404b6a3971d153f6b6699479 |
AAS-2026-10-14-nao-f9d36a4b7276-01 |
2026-10-14 |
code-grounded |
| Sentry Skills |
Sentry Skills supplies host-interpreted workflows, bounded helper checks and evidence-directed guidance revision; execution, enforcement and improvement remain unobserved. |
d18b7aa8ba878354e5c348310230e652f7690f9c |
AAS-2026-10-14-skills-2719567f917c-01 |
2026-10-14 |
code-grounded |
| Graphiti |
Graphiti transforms episodes into temporal graph memory and supplies selected facts to later models; truth, complete retraction and improved capacity remain unestablished. |
964bb11d1e7020e3cc67fab244268b0736b6c219 |
AAS-2026-10-12-graphiti-1d535ca565a4-01 |
2026-10-12 |
code-grounded |
| Phantom |
Phantom wraps SDK-driven work with scheduled execution, selective memory and trace-fed instruction revision; useful operation and improvement remain unobserved. |
f8c7ab42d885936ee54abc785528000260f4acc5 |
AAS-2026-10-15-phantom-cc566f03ac40-01 |
2026-10-15 |
code-grounded |
| harness2 |
Harness² edits and executes agent harnesses, selects outputs and retains trace-derived experience for later improvers; implementation supports the loop, not demonstrated improvement or lesson validity. |
1cfe254e8d25b15f37048c3730134686abe60335 |
AAS-2026-10-22-harness2-d5a180ba559c-01 |
2026-10-22 |
code-grounded |
| ReasoningBank |
ReasoningBank automatically turns task trajectories into retained advice and supplies it to later agents; operation, transfer and improvement remain unobserved. |
ed80611788292ea739f1effd31f16c53823b8a0d |
AAS-2026-10-14-reasoning-bank-f0bedcab086c-01 |
2026-10-14 |
code-grounded |
| link |
Link retains editable Markdown and supplies bounded context through recall and session hooks, but approval, withdrawal and factual warrant remain route-specific. |
cc92f3632121b04063b24eaaf3a8c5429eaf2b57 |
AAS-2026-10-14-link-595f6f9cba36-01 |
2026-10-14 |
code-grounded |
| autocontext |
Autocontext wires evaluated attempts into retained guidance, source revisions and model selection, with route-specific admission but no inspected evidence of exercised improvement. |
f72c154152397f54d55407508b153d28861888a5 |
AAS-2026-10-15-autocontext-b7f59f8405c4-01 |
2026-10-15 |
code-grounded |
| Ultimate |
Ultimate coordinates durable coding work and feeds landings into selective project memory, with path-local checks but no supplied evidence of reliable recall or improved future capacity. |
e2b9fa61962f6aecd394dea1c670688fac3ec432 |
AAS-2026-11-01-ultimate-f5179584de71-01 |
2026-11-01 |
code-grounded |
| PrimeScientist |
PrimeScientist uses benchmark scores and retained experiment trees to select coding attempts and propose successor plans; execution, scientific warrant and improvement remain unestablished. |
29971beac6f4f4b41309b1326762e1b83ceece98 |
AAS-2026-10-14-primescientist-f880686cf98a-01 |
2026-10-14 |
code-grounded |
| MemWiki |
MemWiki scaffolds a project wiki and asks coding agents to maintain and reload it; file writes are implemented, but compliance, accepted knowledge and cross-session benefit are unobserved. |
8034a3da991ac2639b87875172a9572903ecf1d5 |
AAS-2026-10-14-memwiki-79dcc4d04bcb-01 |
2026-10-14 |
code-grounded |
| xMemory |
xMemory retains conversation traces and derived memory for hierarchical, score-guided QA; source wiring supports later reuse, not verified fidelity, uncertainty reduction or improvement. |
1a4f9dd8abd7f2da86bfe6afe4293a1ea1c63608 |
AAS-2026-10-15-xmemory-b9a2b9f58eda-01 |
2026-10-15 |
code-grounded |
| CORAL |
CORAL orchestrates coding experiments through committed candidates, task graders and shared island knowledge; knowledge revision is model-mediated and improvement remains unobserved. |
0123dfb939b35228cf2c1fde224cd0561e727408 |
AAS-2026-10-15-coral-25e5fdab9dc4-01 |
2026-10-15 |
code-grounded |
| academic-research-skills |
Academic-research-skills coordinates research and manuscript work through model roles, retained artifacts and local checks, without demonstrated end-to-end compliance or scholarly accuracy. |
060cb3cbffc2ca1bdfb8f7c6d2ffa4883cf09b1a |
AAS-2026-10-17-academic-research-skills-6fe859d62c16-01 |
2026-10-17 |
code-grounded |
| ModularRSI |
ModularRSI turns task traces into proposed harness edits and later executable generations; admission is wired, but exercised improvement and transfer remain unestablished. |
b5c72c36b0d08ff93f00ee202a8fbdebe849dfb9 |
AAS-2026-10-18-modularrsi-4218ceff5530-01 |
2026-10-18 |
code-grounded |
| AI Research OS Workshop |
AI Research OS Workshop turns selected sources into a persistent, progressively queried filesystem wiki; prompt-mediated admission and structural lint do not establish factual acceptance or improved capacity. |
dc6660555266239cd36483cf51b5ab39febacf02 |
AAS-2026-10-12-ai-research-os-workshop-1a93869f8e7f-01 |
2026-10-12 |
code-grounded |
| o-o |
o-o couples browser-readable research files to Bash-driven Claude updates and document-local reuse, without demonstrated research quality or enforced edit acceptance. |
b0d5063e37b4eafdbc23c1899a31f1836168b989 |
AAS-2026-10-15-o-o-b6fd0055e5c1-01 |
2026-10-15 |
code-grounded |
| Instinctual Memory |
Instinctual Memory turns adapted sessions into Git-backed standing facts and later agent context; structural gates do not establish truth, bound human approval or demonstrate benefit. |
6acb13dc35765bf5ccfc87e445dd09c480f1c28a |
AAS-2026-10-12-instinctual-memory-af48437aa60c-01 |
2026-10-12 |
code-grounded |
| jaz-evals |
jaz-evals runs benchmark agents and carries trace-conditioned advice into later tasks, but external runtime gaps and missing runs prevent claims of exercised improvement. |
2ffe7b5e6792a76fccb26f9ce188c88fb95ffa72 |
AAS-2026-10-15-jaz-evals-17ab95f93794-01 |
2026-10-15 |
code-grounded |
| DocMason |
DocMason pairs file-backed evidence and interaction memory with host-agent workflows and checked answer closure; provenance and admission do not establish semantic truth or improved capacity. |
f3970939593194a158ab7ea1d49995f06cbccc86 |
AAS-2026-10-13-docmason-deba9ae199d5-01 |
2026-10-13 |
code-grounded |
| ReframeWeb |
ReframeWeb couples model-routed tasks with host-owned tools, reviewed guidance and checkpoint-backed startup changes; implementation supports these paths, not demonstrated correctness or improvement. |
318987dcde9a016f8e3db819ed9a73abf6b3ad81 |
AAS-2026-10-14-reframeweb-cff918d7eb1e-01 |
2026-10-14 |
code-grounded |
| Atomic |
Atomic combines retained notes, derived retrieval state and bounded research agents; its publication gates check structure and lineage, not factual support or demonstrated improvement. |
983bea6baf449d060514c94b3b15b5f85f77ba9b |
AAS-2026-10-12-atomic-f98683232831-01 |
2026-10-12 |
code-grounded |
| llm-wiki |
LLM Wiki persists model-extracted claims for later merge, search and synthesis; factual acceptance and candidate-prompt installation are not established. |
d2bd485921f6c7b6071afbb389f44ab8e4e83c69 |
AAS-2026-10-17-llm-wiki-507fade8a56e-01 |
2026-10-17 |
code-grounded |
| kgai |
kgai retains engineering choices and reasons in an event-backed graph, with selective later recall and optional automatic capture but no demonstrated behavioral benefit. |
0ac7f1fae4528805c1d5f504c1bfd6cc34ac953b |
AAS-2026-10-14-kgai-54bcae52e039-01 |
2026-10-14 |
code-grounded |
| MerchantBench |
MerchantBench couples model-selected store actions to delayed simulated feedback and optional run-local notes; source wiring establishes neither exercised learning nor performance benefit. |
d9e5bd7519ab100193c21778aecfb59e30f24651 |
AAS-2026-10-15-merchantbench-d2f4d78dbe37-01 |
2026-10-15 |
code-grounded |
| Synto |
Synto compiles source notes into reusable wiki and pack context with configurable review, durable rejection guidance and structural checks, without demonstrated factual accuracy or improvement. |
91ac6a3681df0552b6c9baef159439c7ebd780c9 |
AAS-2026-10-15-synto-eabcdcce3da2-01 |
2026-10-15 |
code-grounded |
| SAGE |
SAGE combines governed agent memory, optional evidence-support judging and later context delivery; source wiring establishes feedback paths, not useful learning or universal semantic validation. |
e2e0242a421afbd356ed135d7d754a35b0ad5f14 |
AAS-2026-10-15-sage-d3843de6d2dd-01 |
2026-10-15 |
code-grounded |
| SwarmWorld |
SwarmWorld couples local model-agent research, retained experience and bounded artifact programs to an authoritative simulator; inspected wiring does not establish scientific benefit. |
6af7ae9fa36d98b07b0492cf139658e8af1f6eab |
AAS-2026-10-15-swarmworld-e39fbc17e3fb-01 |
2026-10-15 |
code-grounded |
| OpenWiki |
OpenWiki delegates knowledge authoring to models or coding hosts while code maintains evidence debt, durable pages and selective retrieval; semantic accuracy and improvement remain unobserved. |
7c469e59584bfac3f97f5c32a67604de46805999 |
AAS-2026-10-14-openwiki-8f502f680f75-01 |
2026-10-14 |
code-grounded |
| Eidetic |
Eidetic turns session text into durable cards and supplies push context or requested recall; admission and maintenance checks are path-local, with activation and benefit unobserved. |
67506b62fdd0944c243537c3cb10938f7343b6b2 |
AAS-2026-10-14-eidetic-0e54f3c114b3-01 |
2026-10-14 |
code-grounded |
| ExpeL |
ExpeL turns benchmark experience into retained retry guidance and revised rules, then supplies rules and retrieved demonstrations to acting models; operation and improvement remain unobserved. |
e41ec9a24823e7b560c561ab191441b56d9bcefc |
AAS-2026-10-14-expel-efd908f05069-01 |
2026-10-14 |
code-grounded |
| context-constitution |
Context Constitution proposes experience-driven prompt revision, indexed retrieval and compaction, but ships policy documents rather than an executing learning loop. |
964f30137ace9eade1aef112992ed6539b61bbb0 |
AAS-2026-10-15-context-constitution-7a352d7408f6-01 |
2026-10-15 |
doc-grounded |
| Letta Code |
Letta Code couples provider/tool turns to Git-backed memory, transcript reflection and evaluated mod selection; source wiring does not establish semantic accuracy or improved future capacity. |
271f119dcf37d58649a18e6c733386b3c710a456 |
AAS-2026-10-12-letta-code-be76afeab9b7-01 |
2026-10-12 |
code-grounded |
| SkillNote |
SkillNote distributes versioned procedural guidance and uses retained feedback to order candidate skills; host compliance, complete withdrawal and improved task capacity remain unestablished. |
7ea2fc7a8435af64d1c080d8765c0eabba1121ea |
AAS-2026-10-14-skillnote-0ccfe85cda4a-01 |
2026-10-14 |
code-grounded |
| dense-mem |
Dense-Mem combines evidence-linked semantic admission, durable recovery and bounded recall; ordinary evidence recall need not enforce its claimed relationship-support gate. |
bd736450454e237f1ab6d138f4ae9755916d73f6 |
AAS-2026-10-15-dense-mem-1a314bed9a5c-01 |
2026-10-15 |
code-grounded |
| pi-self-learning |
pi-self-learning wires session reflections into persistent rules and later agent context, but checks structure rather than truth and has no observed prevention benefit. |
e0d76a93195b11b688ca71da8a8b5700c6541542 |
AAS-2026-10-15-pi-self-learning-1a318610d4b1-01 |
2026-10-15 |
code-grounded |
| Mem0 |
Mem0 turns conversations into retained facts and supplies selected text to later models; inspected routes establish memory wiring, not truthful extraction or improved answers. |
0724291bd864cf807e88ee8981c8226644e8fb42 |
AAS-2026-10-14-mem0-5b642a1731b4-01 |
2026-10-14 |
code-grounded |
| Memori |
Memori captures interactions and supplies retained facts, summaries and history to later model calls; hosted derivation, successful recall and improvement remain unverified. |
574b1ea3e876f100ef82c37817d603eb7e258e59 |
AAS-2026-10-17-memori-fd4a0418c09b-01 |
2026-10-17 |
code-grounded |
| Napkin |
Napkin combines file-backed Markdown retrieval with external-agent retention and upkeep policies; successful distillation, automatic host delivery and benchmark benefit remain unverified. |
7582d6a46f5a11995956e60a59c41a5b242109f1 |
AAS-2026-10-12-napkin-2b85bef38252-01 |
2026-10-12 |
code-grounded |
| KBLaM |
KBLaM supplies encoded knowledge to transformer attention and trains reusable adapters, but frozen driver defects and source-only evidence limit execution and improvement claims. |
69683ad096e022b58854fd88c0c58968c7d8c470 |
AAS-2026-10-15-kblam-5a0c2fe96cc1-01 |
2026-10-15 |
code-grounded |
| MemPalace |
MemPalace captures and transforms conversation memory for selected later recall and delegated work; source wiring does not establish complete verbatim recall, truth acceptance or improved capacity. |
1392e1810c3ca67776bd7fe3e198788a96dbd6d1 |
AAS-2026-10-14-mempalace-ed71875aea5a-01 |
2026-10-14 |
code-grounded |
| Voyager |
Voyager retains task-success-conditioned Minecraft programs for later prompting and execution; source wiring establishes adaptation, not demonstrated learning gain or reliable transfer. |
55e45a880755d0c8c66ca7fb5fe7962ac8974f89 |
AAS-2026-10-14-voyager-f16a44a0cfe3-01 |
2026-10-14 |
code-grounded |
| Meta-n |
Meta-n turns evaluated solver traces into retained executable interventions, with score-based selection and resume; exercised improvement and generalization remain unestablished. |
b7081843d3c7b0e0f418ca10aaf2ccbff856e7f8 |
AAS-2026-10-14-meta-n-18c23b6f8bb4-01 |
2026-10-14 |
code-grounded |
| EchoWiki |
EchoWiki compiles notes into linked Markdown and reuses retained pages in later model calls, with structural admission rather than factual acceptance or demonstrated improvement. |
11446997423e88177dba62d6a3d5c1e8b4886c62 |
AAS-2026-10-14-echowiki-8d428f79523e-01 |
2026-10-14 |
code-grounded |
| CQ |
CQ retains shared agent advice with local fallback, remote review and feedback-driven retrieval; host verification is requested, while reliance and improvement remain unobserved. |
5765494a3840e5f5a7223c42101f41e85cb9d898 |
AAS-2026-10-14-cq-8fb3dc41c334-01 |
2026-10-14 |
code-grounded |
| binder |
Binder combines typed persistent graph state with selected journal and task read-back; semantic review remains host-mediated and improvement is unobserved. |
e4066d750ac5c6918706f44b12daa261b261106e |
AAS-2026-10-15-binder-a35c41531d40-01 |
2026-10-15 |
code-grounded |
| Agent Skills for Context Engineering |
Agent Skills for Context Engineering couples selectively loaded guidance with file-based research maintenance and trace-fed prompt revision, without established end-to-end operation or improvement. |
9ff0c99108b36d3b886defd825d8ac1f47dba374 |
AAS-2026-10-19-agent-skills-for-context-engineering-7e663338370b-01 |
2026-10-19 |
code-grounded |
| auto-harness |
auto-harness couples coding-agent revisions to benchmark gates and persistent evaluation controls; source establishes standing feedback paths, not exercised improvement or enforced isolation. |
52360dbbeb8cea8886d90778e634a145c51da735 |
AAS-2026-10-15-auto-harness-92303e444cb2-01 |
2026-10-15 |
code-grounded |
| nuggets |
Nuggets retains readable facts, reconstructs HRR recall and supplies later Pi context and scheduled prompts; truth acceptance, reliable delivery and improvement remain unestablished. |
714cab8a3b1fb843aa98dfb51584d2c07a6739f3 |
AAS-2026-10-14-nuggets-c2e87e6916cc-01 |
2026-10-14 |
code-grounded |
| wuphf |
Wuphf coordinates office agents and saved executable routines with trace-fed knowledge reuse, but approval, verification and recovery guarantees remain route-local. |
88023d4dac7491b6e71dbc1e86d274c435b9cb2c |
AAS-2026-10-15-wuphf-b246722f99c0-01 |
2026-10-15 |
code-grounded |
| amanuensis |
Amanuensis couples agent-led codebase surveys to revision-bound memory and bounded readers; its gates control record authority, not source truth or demonstrated future improvement. |
a98093dc7aad3c0e64d42132ece305fc5ea7f3f9 |
AAS-2026-11-08-amanuensis-b3a8e3e53488-01 |
2026-11-08 |
code-grounded |
| QwenPaw |
QwenPaw orchestrates personal-assistant work with backend-dependent memory, recoverable context and forked workers; wiring establishes neither answer truth nor improved capacity. |
16b2e712d198166c9359e77055b13bf2668bf5ae |
AAS-2026-10-15-qwenpaw-e41495d47fc0-01 |
2026-10-15 |
code-grounded |
| secure-llm-wiki |
Secure LLM-Wiki compiles sources into gated claims for later model queries; security admission and evidence-labelled delivery do not establish factual warrant or injection resistance. |
dfbc60e37487b61e9a5cbce1271a13576a776d51 |
AAS-2026-10-19-secure-llm-wiki-df02f419bddb-01 |
2026-10-19 |
code-grounded |
| Reflexion |
Reflexion wires task-local feedback into retained retry guidance, but delivery differs across workflows and neither diagnosis validity nor improvement is demonstrated. |
218cf0ef1df84b05ce379dd4a8e47f17766733a0 |
AAS-2026-10-14-reflexion-19159b69b712-01 |
2026-10-14 |
code-grounded |
| IsaacCLupus_mnemosyn_spec |
Mnemosyne proposes audited compilation and retained-feedback memory; its legacy bridge wires publication and retrieval without establishing the unified lifecycle's guarantees. |
0fda14eb33d2fc01a4e3969a9ab487b22c4f9d92 |
AAS-2026-10-15-isaacclupus-mnemosyn-spec-1887e13dcc13-01 |
2026-10-15 |
doc-grounded |
| Hermes Agent |
Hermes Agent combines model-directed tool work with trace-fed notes, skills and continuation handoffs; admission and recall are route-specific, and improved future capacity is unestablished. |
0e427d788d253630d38a3a51cd76849e73cd6925 |
AAS-2026-10-14-hermes-agent-bfb4bdc54d37-01 |
2026-10-14 |
code-grounded |
| llm-wiki |
llm-wiki combines host-model compilation and indexed Markdown retrieval with executable session memory and bounded checks; source evidence does not establish reliable operation or improvement. |
95a042c68998559305bc69ac8c482c4e1a415ad3 |
AAS-2026-10-17-llm-wiki-a93f86f698d7-01 |
2026-10-17 |
code-grounded |
| SoL-Pi |
SoL-Pi fuses mutation and commands, archives observations, selects diagnostic quotes and schedules native compaction; wiring is supported, but evidence completeness and realized benefits are not. |
41fa7e0e230bedc5e639aed2e7b9cc5698726db1 |
AAS-2026-10-14-sol-pi-42f04fbbc6bb-01 |
2026-10-14 |
code-grounded |
| semantic-engine |
Semantic Engine uses operator-run ingestion and retained chunk embeddings to retrieve full documents; factual acceptance, retrieval benefit and connected visualization remain unestablished. |
7e2481671cad01f0b4c89f2a1877d6733d7e552c |
AAS-2026-10-17-semantic-engine-666c064cd8dc-01 |
2026-10-17 |
code-grounded |
| OpenSage |
OpenSage lets models construct agents and scripts, retain file knowledge and compress history; implemented coordination and checking do not establish exercised self-improvement. |
54d64702e18d195c8f0fed711c6db5dc1d62ab29 |
AAS-2026-10-14-opensage-291d1ed79d37-01 |
2026-10-14 |
code-grounded |
| OS-Copilot |
OS-Copilot links model planning, host execution and feedback-gated tool memory; repair and later reuse are wired, but task success and exercised self-improvement remain unobserved. |
f720af8807e49a92dda64572d2c6bc6c0ac7ee7e |
AAS-2026-10-16-os-copilot-7cef03148c53-01 |
2026-10-16 |
code-grounded |
| HippoRAG |
HippoRAG retains passages and model-derived graph/vector search state for later QA; structural checks and relevance filtering do not establish factual acceptance or measured improvement. |
d5c8329422e0a0b834a15874545cb6a74b4f9b26 |
AAS-2026-10-16-hipporag-2b37edfc7ade-01 |
2026-10-16 |
code-grounded |
| SkillWeaver |
SkillWeaver turns browser practice into retained Python skills and later selection policy; the update-to-consumer path is implemented, but exercised improvement and generalized correctness are unestablished. |
f2a63d65d0f6ff46ac30e817cede8797f8f25b97 |
AAS-2026-10-14-skillweaver-c921cc7a5ed0-01 |
2026-10-14 |
code-grounded |
| Enoch |
Enoch turns conversation and task history into retained context and improvement proposals, with bounded validation and staged review/update controls but no observed self-improvement. |
0934ad59bd253adfffb8263c65e4bf0f4e2d7a9f |
AAS-2026-10-13-enoch-a73695a57e3b-01 |
2026-10-13 |
code-grounded |
| llm-wiki |
llm-wiki retains session-derived Markdown for later agent queries, with deferred foreground synthesis and optional review rather than universal semantic acceptance. |
b1088890ee0743810a92577aecad946c6b3eb2d2 |
AAS-2026-10-16-llm-wiki-6786c5985cdc-01 |
2026-10-16 |
code-grounded |
| prime-agent |
Prime Agent combines a Rust model/tool loop with persistent Python, bounded child orchestration and trajectory-fed harness editing; later delivery is wired, but improvement is unobserved. |
33da15a2c15c4e903c372f69ae8e1156d77d53da |
AAS-2026-10-14-prime-agent-8ec59ae12476-01 |
2026-10-14 |
code-grounded |
| Prove2Me_workspace |
Prove2Me_workspace directs source-faithful Lean formalization, proof checking and reuse through distinct review gates; its host, service enforcement and learning remain unobserved. |
38e12f9d215cb103a0f9173ec230a614fcd93b61 |
AAS-2026-10-14-prove2me-workspace-187ef30582c9-01 |
2026-10-14 |
doc-grounded |
| echoes-vault-opencode |
EchoesVault restores repository-local Markdown through generated discovery and recent notes; its shared writer checks structure and concurrency, not project truth or learning benefit. |
875c45aa45559529b87a7481b0c21e6dfd8c5eb0 |
AAS-2026-10-15-echoes-vault-opencode-239c6738bb9e-01 |
2026-10-15 |
code-grounded |
| okf-harness |
OKF Harness couples agent-maintained Markdown knowledge with deterministic lineage checks and bounded retrieval; semantic correctness and instruction compliance remain agent responsibilities. |
0dd4f806c454a68088c5a3e64b4cbebbe2c216be |
AAS-2026-10-15-okf-harness-6bf5507fbdbf-01 |
2026-10-15 |
code-grounded |
| Hyalo |
Hyalo combines local Markdown retrieval and mutation with agent-directed maintenance and optional Jev suggestions; structural gates do not establish semantic correctness or exercised improvement. |
697f215fe8a90a5c760fceb686f2a321c9db7221 |
AAS-2026-10-14-hyalo-ffe11d6a6d56-01 |
2026-10-14 |
code-grounded |
| Tolaria |
Tolaria supplies editable Markdown, bounded conversation replay and agent-accessible vault tools; inspected admission checks govern file operations, not factual acceptance. |
48f2a70844c1e8005a7aa5f1499c66f7678360de |
AAS-2026-10-14-tolaria-42ddaf2ee8cd-01 |
2026-10-14 |
code-grounded |
| Thalo |
Thalo combines file-backed knowledge, structural feedback and incremental model synthesis, with recurrent prose read-back but incomplete invalidation and no demonstrated semantic acceptance. |
cdb9aae983e6bc0b75eff1606bc99b088c3aebff |
AAS-2026-10-14-thalo-86fcc3efe983-01 |
2026-10-14 |
code-grounded |
| talorys |
Talorys combines personal-information tools, retained facts and summaries, optional semantic recall and owner-approved destructive actions; operation and benefit remain unobserved. |
c5da440010402ed4c1b9789ae22eab96a8f0f379 |
AAS-2026-10-15-talorys-7bce41ba9401-01 |
2026-10-15 |
code-grounded |
| EvoOntology |
EvoOntology retains versioned semantic guidance and gates candidate publication on recorded comparisons; host grounding, evaluation integrity and successful improvement remain unobserved. |
c94389b764394379086ff25e50d0e0f4396167b4 |
AAS-2026-10-13-evoontology-f2d0c7992c6e-01 |
2026-10-13 |
code-grounded |
| REM |
REM combines episodic storage, model-proposed facts and ranked context retrieval, but shipped maintenance and SDK handoffs do not complete the declared memory loop. |
935e8be0a1fca5b23dbabfe16c48b562c1cd24cc |
AAS-2026-10-14-rem-a81fdc774400-01 |
2026-10-14 |
code-grounded |
| equipa |
EQUIPA orchestrates developer/tester work, gated Git delivery and trace-fed context and prompt changes; source wiring does not establish exercised improvement or universal test coverage. |
fa8f41a6948584ee53e804edbb08b7449d05cd14 |
AAS-2026-10-15-equipa-32016e8dd9ae-01 |
2026-10-15 |
code-grounded |
| cludebot |
Cludebot turns retained exchanges and outcomes into later model context and procedural guidance; source wiring establishes adaptation pathways, not exercised improvement or verified belief correction. |
cecf234f10397b1e21dec2ff65ded864982f3564 |
AAS-2026-10-15-cludebot-cb70d11d215b-01 |
2026-10-15 |
code-grounded |
| Tendril |
Tendril retains model-authored TypeScript for later execution; file-mediated self-extension is implemented, but repair and cross-session improvement remain unobserved. |
e671a4143d28de68289efd81580002041bb4cb6a |
AAS-2026-10-15-tendril-f560fb299a37-01 |
2026-10-15 |
code-grounded |
| interview-doc-agent |
Interview Doc Agent organizes personal records into reusable dossiers and selects them for job documents through host-interpreted instructions, without demonstrated factual checking or execution guarantees. |
6ebfe9aba989d7949dadea0fc4af6fb64892c6aa |
AAS-2026-10-14-interview-doc-agent-c7597c33f23d-01 |
2026-10-14 |
doc-grounded |
| sig-releases |
Sig declares local work-note synthesis, human-approved team sharing and later AI context; its release documentation does not establish enforcement, context selection or improved answers. |
314bbea9d1cef911cea0f201eec66a57c2e59020 |
AAS-2026-10-14-sig-releases-50379145ac31-01 |
2026-10-14 |
doc-grounded |
| Signet |
Signet captures evidence, maintains source-linked memory and supplies selected context; provenance gates and durable feedback updates do not establish truth or improved performance. |
12dba1ce753babda735f2ba7b6f150c6c0b7f95e |
AAS-2026-10-14-signetai-64f703d83203-01 |
2026-10-14 |
code-grounded |
| Agent S |
Agent S closes a model-driven desktop loop, with optional coding, historical experience reuse and offline rollout selection; verification is mostly policy and improvement remains unobserved. |
15776a8dc659658b1963816ea006f7af1978e595 |
AAS-2026-10-16-agent-s-db6fee212353-01 |
2026-10-16 |
code-grounded |
| gnosis |
Gnosis retains repository rationale in JSONL and returns lexical search results to coding agents; semantic admission and later use remain caller-owned and unobserved. |
429528bf98e424b40837d2dbe3631d0fb42da6fd |
AAS-2026-10-15-gnosis-5bfb42422ce3-01 |
2026-10-15 |
code-grounded |
| AutoSci |
AutoSci combines prompted research with a persistent wiki and bounded experimental search; scientific acceptance is procedural, while generated-payload transport remains conditional. |
f5c2fe009c6683b6aa5b668de759a0610c938aaf |
AAS-2026-10-15-autosci-430df584df8f-01 |
2026-10-15 |
code-grounded |
| Spacebot |
Spacebot encloses conversational and autonomous work with retained memory and procedure revision, but source wiring does not establish correct delivery or beneficial self-improvement. |
5c2ee47b397e09c956dc15000ea576ecd28e92bb |
AAS-2026-10-14-spacebot-6a954b400f95-01 |
2026-10-14 |
code-grounded |
| WikiSkill |
WikiSkill turns retained task experience into Wiki lessons and evaluator-gated instructions; delivery is implemented, but exercised improvement and explanatory warrant remain unestablished. |
da6984caca09d5f81b2336cf3067b59c0c18bb1e |
AAS-2026-10-15-wikiskill-baafad05d35e-01 |
2026-10-15 |
code-grounded |
| meta-harness |
Meta-Harness searches executable model harnesses using retained evaluation feedback; source inspection establishes adaptation pathways, not exercised improvement or causal mechanism effects. |
8123ccabe2b19fa1123090b2e5f1bacc20963ce8 |
AAS-2026-10-14-meta-harness-5e480e270fee-01 |
2026-10-14 |
code-grounded |
| Company Brain |
Company Brain combines a Slack model/tool runtime with hosted company memory, learned style and reviewed research drafts; local controls establish wiring, not factual warrant or demonstrated improvement. |
ef8a45ef3aed97b32599c7d8007edefab6d14e9a |
AAS-2026-10-14-company-brain-30a9e8cfaeca-01 |
2026-10-14 |
code-grounded |
| supermemory |
Supermemory's public adapters submit conversations and supply retained context; extraction, review and graph maintenance remain declared contracts of an opaque engine. |
f81fad00e4bfc42610159cf6bd7bdb5fc4ca7cbe |
AAS-2026-10-19-supermemory-1cc18c778c48-01 |
2026-10-19 |
code-grounded |
| Dynamic Cheatsheet |
Dynamic Cheatsheet carries model-curated reference text and retrieved responses into later solving without a semantic admission gate or established causal benefit. |
5cfe3c37e8e52b1d858d0f3df46e7f17c50991b9 |
AAS-2026-10-12-dynamic-cheatsheet-ca0a0b29d273-01 |
2026-10-12 |
code-grounded |
| Swamp |
Swamp couples external agents to typed automation, retained operational state and review gates; implemented checks do not establish semantic accuracy or exercised self-improvement. |
380e3c307164501c7d705f7707a864f4612b2785 |
AAS-2026-10-15-swamp-440d19cc0bcc-01 |
2026-10-15 |
code-grounded |
| WeKnora |
WeKnora connects indexed documents, editable wiki pages and trace-fed personalization to RAG and agent work, with bounded write gates but no observed benefit or universal truth check. |
234aa948db29c10f5ed4b7939285c7b3711c5d4c |
AAS-2026-10-15-weknora-7adf8a01c381-01 |
2026-10-15 |
code-grounded |
| Pond |
Pond archives agent sessions for caller-directed recall and fidelity-limited restoration; derived retrieval updates are wired, but host adoption and task benefit remain unobserved. |
0c440e130faf44052d0d896a585e572090079f1f |
AAS-2026-10-14-pond-7a589a889ca1-01 |
2026-10-14 |
code-grounded |
| Semiont |
Semiont connects event-backed knowledge and selected context to queued model work; admission checks provenance, not truth, and generated-citation commits encounter an assignment mismatch. |
a920067ac4ea63a57b1855b7d263485fbd22bab4 |
AAS-2026-10-15-semiont-d516c7013d8f-01 |
2026-10-15 |
code-grounded |
| ai-modules |
ai-modules combines model-interpreted task and wiki procedures with executable bookkeeping; retained feedback can shape later work, but compliance and improvement remain unobserved. |
0cf0227c416f75a79d7317f603a209a63e7a3c8f |
AAS-2026-10-14-ai-modules-a7044d5627d2-01 |
2026-10-14 |
code-grounded |
| eal-bench |
EAL-Bench evaluates authorization-memory formation and downstream tool decisions, with structural admission and wired outcome feedback but no established live improvement. |
51648690bc52d7a9c7ac080a2c67a784fe9d56cb |
AAS-2026-10-15-eal-bench-f9a76e2fa858-01 |
2026-10-15 |
code-grounded |
| Cognee |
Cognee turns corpus and session material into retrievable graph content, guidance and ranking state, with model-mediated lesson admission but unobserved benefit and partial lifecycle coverage. |
f154b32a9e544dac73e7ddf2010327cf9988edcd |
AAS-2026-10-14-cognee-d8839fe03c5b-01 |
2026-10-14 |
code-grounded |
| Fractal |
Fractal delegates direct workspace coding to PredictRLM and carries session context across turns; typed returns and retained traces do not establish verified edits or improved capacity. |
6a226e7d539e229b5c7cc9ffa3b05ab1c8f00385 |
AAS-2026-10-14-fractal-ee03010af4c2-01 |
2026-10-14 |
code-grounded |
| playground |
Playground exposes persistent tenant shells, bounded job replay and optional native-worker routing; model-memory assembly remains doctrine or external implementation. |
de28ac8fdda30c7a5c2ee099de089e65a685195f |
AAS-2026-10-14-playground-88ee8a292f9b-01 |
2026-10-14 |
code-grounded |
| Kompl |
Kompl compiles sources into a reusable wiki through fixed orchestration and model drafting; admission, clipped retrieval and constrained FAQ adoption do not establish factual improvement. |
17970c969b0c2576bf722840e0e2db3c6416a7aa |
AAS-2026-10-14-kompl-9cffb1e5eb86-01 |
2026-10-14 |
code-grounded |
| funes |
Funes prescribes source-preserving wiki compilation and later cited reuse, with executable extraction and mechanical audits but no observed semantic workflow or improvement. |
3fb448b8e0d1ab5f6de3c6de61d8d556dab4ee11 |
AAS-2026-10-15-funes-a9e6dbb3946a-01 |
2026-10-15 |
code-grounded |
| Continuity |
Continuity combines a desktop model/tool loop with retained facts, narrative context and isolated research; source wiring establishes continuity mechanisms, not factual acceptance or demonstrated improvement. |
4ca8f6b4108aa4494e3861ed33c8019dbd662c67 |
AAS-2026-10-14-continuity-3e490e1d39f0-01 |
2026-10-14 |
code-grounded |
| PageIndex |
PageIndex retains document trees and pages for model-led selective reading; structural admission and citation lookup do not certify answer truth or measured benefit. |
af92d9658a6a9c0ae31171b43198db28951af2f4 |
AAS-2026-10-15-pageindex-50056ef72767-01 |
2026-10-15 |
code-grounded |
| Hindsight |
Hindsight turns submitted experience into revisable facts and standing pages for later model context; source-level wiring does not establish truth or improved capacity. |
610c135b6f2d0d8b972da94146ae635943f36271 |
AAS-2026-10-14-hindsight-34c2ca821edf-01 |
2026-10-14 |
code-grounded |
| VLM-wiki |
VLM-wiki combines instructed media-to-wiki authoring with file-based import, deterministic reviews and optional push delivery; host compliance, factual acceptance and improved capacity remain unestablished. |
503c43f98b5d1e21c74df384fdffd3e0652573b3 |
AAS-2026-10-14-vlm-wiki-311faef777b4-01 |
2026-10-14 |
code-grounded |
| browzy.ai |
browzy.ai compiles sources into a local wiki and supplies lexical retrieval to answer models; structural admission and retained summaries do not establish factual warrant or improved capacity. |
56c253042041ee2f483a5e9b824174d746891cf4 |
AAS-2026-10-15-browzy-ai-0a823a13a7b6-01 |
2026-10-15 |
code-grounded |
| Siftly |
Siftly retains imported bookmarks and AI annotations for classification and retrieval; application code controls progression, while accuracy and improved future performance remain unobserved. |
b25daa45b858f4be096b5c368671c34db4407d8e |
AAS-2026-10-18-siftly-b81c94cf0f77-01 |
2026-10-18 |
code-grounded |
| Virtual Context |
Virtual Context retains conversation-derived memory for bounded context assembly and on-demand reload, with operative correction gates but no demonstrated answer benefit or universal fidelity guarantee. |
74ccc570c40bb551820fcd330c42e0dfc0588f05 |
AAS-2026-10-14-virtual-context-bbd06cc3ac95-01 |
2026-10-14 |
code-grounded |
| Voiden |
Voiden exposes project requests through MCP, carries execution values into later requests and records outcomes; host judgment and plugin verification remain outside the inspected core. |
d7a0d05c40acb6cc483000643c700f17aecba0ef |
AAS-2026-10-14-voiden-98188b4a2f55-01 |
2026-10-14 |
code-grounded |
| OpenViking |
OpenViking turns session traces into retained context and reusable policy files, then selects text for shipped consumers; wiring is supported, truth acceptance and improved capacity are not established. |
5242d28d7792411e1a6eeecdc2b1d94d7fc98e6b |
AAS-2026-10-14-openviking-7b5b57541af5-01 |
2026-10-14 |
code-grounded |
| deja-vu |
deja-vu turns coding-agent history into queried recall, event-selected context and trace-derived suggestions; host compliance, successful repair and improved capacity remain unobserved. |
4c08fb26ad8a8b8e656a120f06ca4175a7078148 |
AAS-2026-10-14-deja-vu-22137d4d39a5-01 |
2026-10-14 |
code-grounded |
| LLM-Wiki-v3 |
LLM-Wiki-v3 turns source documents into reviewed Markdown and ranked identifiers; lexical checks and path-specific admission do not establish semantic truth or downstream agent benefit. |
267541db838cbca757a112bd0cfa47fa61eb5fb2 |
AAS-2026-10-18-llm-wiki-v3-0c553a3118eb-01 |
2026-10-18 |
code-grounded |
| SkillLift |
SkillLift revises agent skills and reusable scoring rubrics through benchmark rank feedback; the wired improvement plane has unobserved benefits and a conditional refinement-replay risk. |
599358b4d4c4a27c0e004df8228ab93026600653 |
AAS-2026-10-14-skilllift-0f350bf64418-01 |
2026-10-14 |
code-grounded |
| Auto-claude-code-research-in-sleep |
ARIS coordinates research through host-followed skills, local gates and selective retained context; human-gated harness revision is designed, but execution and improvement remain unobserved. |
26b95cfa0d8747078e9e43b42e20952709e561b8 |
AAS-2026-10-15-auto-claude-code-research-in-sleep-a8abf5f80cbb-01 |
2026-10-15 |
code-grounded |
| memex |
Memex delegates persistent wiki editing and later file search to Claude Code; schema adaptation is requested, while grounding, reliable execution and improvement remain unobserved. |
1e223b6045be31b70b1ab9e4b28dd403150d29e7 |
AAS-2026-10-14-memex-42ecde52cc05-01 |
2026-10-14 |
code-grounded |
| Self-Training-LLM |
Self-Training-LLM wires generated answers and consistency scores into SFT/DPO checkpoints and later inference; operator-mediated stages do not demonstrate factual or learning gains. |
97839b29d0fd8bb474f5549fa3e9d6ca504732e0 |
AAS-2026-10-15-self-training-llm-5b5ee621bde2-01 |
2026-10-15 |
code-grounded |
| claude-code-dynamic-workflows |
Claude Code dynamic workflows put orchestration and intermediate results in JavaScript, with documented claim checking and session replay but no observed reliability or improvement evidence. |
workflows-2026-10-14 |
AAS-2026-10-14-workflows-f9c2f80a3a94-01 |
2026-10-14 |
doc-grounded |
| A-MEM |
A-MEM automatically derives semantic metadata and edits related notes, but split read representations and missing execution evidence limit claims of reliable improvement. |
f303dfc71e07bdc787f4bc135d4cea328ae30e99 |
AAS-2026-10-16-a-mem-sys-eff95cd80440-01 |
2026-10-16 |
code-grounded |
| sage-wiki |
Sage-wiki compiles sources and session text into reusable wiki and graph state, with route-local admission controls but no demonstrated accuracy or improvement. |
d3ef3a1346405c82eac01b4d7c493c94c1c59169 |
AAS-2026-10-15-sage-wiki-cc43456fb833-01 |
2026-10-15 |
code-grounded |
| openclerk |
OpenClerk supplies cited task context and explicit knowledge maintenance over Markdown and SQLite; source-sensitive repair is implemented, but approval and later benefit remain unobserved. |
092fb98d243c2d9d95867ad04618ebce88983552 |
AAS-2026-10-16-openclerk-ec0e00c2bb0a-01 |
2026-10-16 |
code-grounded |
| Sparks |
Sparks couples external-agent wiki writing to deterministic vault maintenance and requested retrieval, without establishing semantic acceptance or improved future capacity. |
49abddc848ea55f78518bfbbbe04c3d37a5f2705 |
AAS-2026-10-15-sparks-563865f8b26f-01 |
2026-10-15 |
code-grounded |
| ai-memex-cli |
ai-memex-cli delegates Markdown knowledge maintenance to coding agents, feeds structural defects into repair, and supplies later context without demonstrated semantic accuracy or improvement. |
4adb274a33cbb9ff0f03a6aeeb9c8d2525a5ac3f |
AAS-2026-10-15-ai-memex-cli-4324549b4bc0-01 |
2026-10-15 |
code-grounded |
| SkillX |
SkillX turns supplied experience into retained plans and reference skills for later prompts; operative filters and retrieval are implemented, but exercised improvement and complete deployment remain unestablished. |
b0047c848dbbcf019363d6d56eef8c558f69aefc |
AAS-2026-10-14-skillx-61e3fc8f9aa5-01 |
2026-10-14 |
code-grounded |
| lhtb |
LHTB couples model-driven terminal work to task-owned verification, budgeted repair and retained context; wiring does not establish successful repair or general scientific validity. |
7ca0cacea9bf6a0694eaad7306bed794eaaaabce |
AAS-2026-10-15-lhtb-a0548ba4edbf-01 |
2026-10-15 |
code-grounded |
| Agent Workflow Memory |
Agent Workflow Memory regenerates file-backed web-task guidance from examples and traces; later model delivery is wired, but workflow validity and beneficial use are unobserved. |
8c0ff8cd11d648c8fceb99e4e42f37e3b75381b1 |
AAS-2026-10-12-agent-workflow-memory-cb2da079273e-01 |
2026-10-12 |
code-grounded |