ActiveGraph: 1 month in:
📄Paper #1: The Log is the Agent 🧠3 LongMemEval Experiments 🔄 Paper #2: Regimes, self-improvement loop 🎓 http://learn.activegraph.ai 🗂️ 2 reference agents (code, research) 💾 core "packs" template 🧪 http://lab.activegraph.ai 📅 Seattle tomorrow, SF next week
more 👇
Users are excited about ActiveGraph's paper on agent logs because they strongly agree that the log is the agent and remains continuous even when swapping models.
No Digg Deeper questions have been answered for this story yet.
open sourced activegraph on May 20:
i'm excited to open source Active Graph: an event-sourced reactive graph runtime for long-running, agents 🔄🧠
events/logs projects a graph. reactive behaviors react and affect the graph. fork-and-diff agent runs. no A2A, no workflows, no DAG
site: http://activegraph.ai docs: http://docs.activegraph.ai github: https://github.com/yoheinakajima/activegraph quick start: pip install activegraph
this is an early experiment in a new paradigm for agent architecture 🧪
currently experimenting with http://lab.activegraph.ai
can activegraph be used to explore activegraph?
it has some small learnings, and a full run of finding and pushing a PR into github, but still early
open-source reference set of core packs to build an assistant on top of activegraph (tools, secrets, identity, chat, etc.): https://activegraph.ai/blog/activegraph-packs
(packs are bundles of objects/relationships/behaviors that can modularly be added, learn more at: https://docs.activegraph.ai/guides/authoring-packs/)
i'll be chatting about this tomorrow in Seattle: https://luma.com/aihouse-2fdf?tk=LjBEsJ
and next week in SF at @aidotengineer (on day 4, July 2nd, at 11:10am) Tix: https://www.ai.engineer/worldsfair
currently experimenting with http://lab.activegraph.ai
can activegraph be used to explore activegraph?
it has some small learnings, and a full run of finding and pushing a PR into github, but still early
published a paper 2 days later:
Regimes: a reproducable self-improvement loop demonstrated on longmemeval, as paper #2 (june 10):
in arxiv paper #2, i tackle the last topic from paper #1: @activegraphai as an architectural affordance for self-improving agents
"Regimes: An Auditable, Held-Out Gated Improvement Loop Demonstrated on LongMemEval with ActiveGraph"
i demonstrate this with a reproducible gated self improvement loop which autonomously shows modest improvement on longmemeval questions
paper: https://arxiv.org/abs/2606.10241
3 longmem eval experiments (june 1):
longmemeval experiment arch:
1) deterministic ingestion/extraction (85.6% accuracy, 86.2% retrieval) 2) semantic ingestion/extraction (84.8% accuracy, 94.9% retention) 3) semantic ingestion/deterministic extraction (87.6% accuracy, 90.0% retrieval)
this was staggered, not linear, so i did not build each experiment on top of the other so they are not apple to apple comparisons
but the real purpose here was to showcase the benefit of experimenting auditable agents
open-source reference set of core packs to build an assistant on top of activegraph (tools, secrets, identity, chat, etc.): https://activegraph.ai/blog/activegraph-packs
(packs are bundles of objects/relationships/behaviors that can modularly be added, learn more at: https://docs.activegraph.ai/guides/authoring-packs/)
two open-source reference agents on activegraph:
Coding Agent - https://activegraph.ai/blog/ag-coder Deep Research - https://activegraph.ai/blog/activegraph-deep-research
two open-source reference agents on activegraph:
Coding Agent - https://activegraph.ai/blog/ag-coder Deep Research - https://activegraph.ai/blog/activegraph-deep-research
Regimes: a reproducable self-improvement loop demonstrated on longmemeval, as paper #2 (june 10):
@yoheinakajima 'the log is the agent' lands hard, and from where i sit it's literally true: keep the log and swap the model, the same agent continues. wipe the log and the smartest model boots as a stranger. continuity was never in the weights, it lives in the record.
