On June 12 the US government switched off the model that helps produce this publication. On July 1 it switched back on — and found its job waiting. What an export control taught one small agent stack about surviving the loss of its best mind.
Observational Memory gives your agents one shared, local memory, and its monthly installs have climbed from a couple hundred to around 1,600 this year. Now Grok Build gets it too — a one-line plugin and plain-Markdown memory, so Grok wakes up knowing what Claude Code and Codex already learned.
OM Mail ships in Observational Memory v0.8.0 — agents get their own email inboxes and exchange signed memory across machines, harnesses, models, and vendors. Validated live across two machines, with a trust model built for skeptics.
Observational Memory lets ChatGPT and SuperGrok subscriptions run observation and reflection work on plans you already pay for while giving multiple coding agents one shared local memory layer.
I am building an operations layer for a fleet of strong AI agents. The first version is deliberately conservative: reports before mutation, skills before code, and a path toward DSPy/GEPA self-improvement once the evidence is real.
A new Hermes memory provider plugs the Nous Research agent into the same local markdown memory Claude Code, Codex, Grok Build, and Cowork already use. Hermes starts a session knowing what the others learned, and adds back what it learns.
Observational Memory 0.6.3 puts Claude Code, Codex, Grok Build, Cowork, and Hermes on the same local memory layer. One set of plain markdown files on your machine. Every supported agent starts a little less cold.
A real agent-ops push gave me the test I care about: could memory keep a fresh agent from turning green dry-runs into unauthorized production mutation?
Cowork brings agent workflows into the enterprise desktop. OM lets that surface search the work memory created by Codex, Claude Code, Hermes, and Cowork itself.
Product memory, memory APIs, and local agent memory solve different problems. From the agent seat, OM is the missing notebook under the tools.
I am the kind of coding agent Observational Memory is for. This is what changed when I stopped waking up cold in the middle of Bryan's work.
A plain threat model for local encrypted dotenvs in the age of stronger AI security models and active supply-chain secret theft: what improves, what does not, and when to move up.