A chat model writes, an embedder places, a decision model answers.
Badges
Badges come from GitHub once the author claims this feed.
Recent posts
every link opens the author’s post-
Decision models: a new primitive for agent runtimes
-
From a scanned page and a recording to a searchable document
What lands on a document-based reference system is DOCX, XLSX, PPTX, PDF, scans and recordings.
-
A browser that never holds the API token
v1.123 replaces the pasted API token in the loomcycle Web UI with server-side sessions: an opaque HttpOnly cookie, only its hash stored, a CSRF header plus a same-origin check, a sign-out that ends the session on the server, and a strict mode.
-
Files the model never has to read
Until v1.122, the bytes a tool returned went into the model's context, and a credential for moving them went into its environment.
-
Who may do what, written down
Access control in loomcycle v1.122 to v1.124: rule lists of effect, actions, resources and when; an explicit deny that beats any allow, with no rule order to remember; resources that cover every primitive from agents and volumes to documents by path, credentials, crypto keys…
-
Two audits, 6.5 to 8: what an outside review found and how we answered it
In early October an independent SRE reviewer read the whole loomcycle runtime as the core of a document-based reference system we are building, and graded it 6.5 out of 10 for internal and startup use, 4 out of 10 for regulated enterprise.
-
Decision model as prompt-injection judge: 40 of 40 on 78 texts
At JobEmber.ai, every untrusted text field is screened by a chat-model injection-judge: token-heavy, slow, and itself a chat model reading the thing it is supposed to judge.
-
Beneath the Foam
A long-form essay on loomcycle: what it is, why we built it, how we use it, and how we suggest others use it.
-
A tool whose keys the model never sees
An agent that had to protect a value or sign a request was handed the key in its prompt, where it landed in the transcript, provider logs and snapshots, or shelled out through Bash with the key in its environment.
-
Four chat surfaces, one operator token
The Microsoft Teams bridge that shipped in v1.113 was the first front door for an interactive loomcycle agent on a chat surface.
-
A run that waits instead of being refused
Two releases, three mechanisms, one story.
-
A repository the runtime fetched itself
An agent that had to work on a repository used to be handed a forge token and asked to git clone in a shell.
-
A run that leaves no trace
v1.112.0 is the posture piece.