Library
Every file your agents need, in one library.
The library holds every file your coding agents work from, in any format, and record types for the files that carry a purpose. It's where the context layer keeps them: written once, opened in a real editor, and read by every agent and everyone on the project.
Files in any format
A file is anything your project keeps, in its own format, from Markdown, JSON and YAML to images and PDFs: the context files for AI your agents read and write, in one place every agent can reach.
Folders and tags from your terms help people and agents find each file. What your agents write lands beside yours, and a file only you wrote can stay private.
Record types: files with a purpose
A record type gives a kind of file a meaning your agents can rely on: one format, the fields it carries and, where it needs one, a lifecycle of statuses.
Define one wherever a kind of file needs it, such as a bug with its severity. Every record of the type reads the same, your agents read its fields instead of guessing, and a board shows its statuses as columns.
Artifacts for your workflows
An artifact is a file a workflow uses: as the input a node reads, as the output a node writes, or as the template an output must follow.
Each agent starts its task with its inputs in hand. A node finishes only once its outputs exist, each in its template's shape, so the next node never reads a half-finished file.
One set of rules for every agent
A rule is a record your agents follow, written once in the library and read by Claude Code, Codex, Cursor, Gemini CLI and Grok alike.
Mix agents per task, and each works from the same rules. A task starts with only the rules that apply to it, so no agent spends tokens reading the rest.
Versions and history
Every save makes a new version, and every change is recorded: who made it, when and why.
Compare any two versions and restore one when an edit went wrong. When two people or agents change the same lines, you choose what stays.
Plain files, always yours
Every file keeps its own name and format, so what you keep in the library reads the same outside ctxflows.
Export everything at any time: Markdown stays Markdown, JSON stays JSON, and a record keeps its fields.
A triage that comes out in the same shape every time
Your error reports arrive in every form. You want each triage to come out in one shape, whichever agent writes it.
Add the template
You put triage.schema.json in the library: severity, area, the steps to reproduce and a suspected cause.
Name it in the workflow
Your triage workflow's Triage node reads the error report and the rule “How we rate severity”, and its output follows the schema.
The agent writes the output
Claude Code reads both and writes triage.json. An output without a suspected cause is refused, and the node stays open until the file is complete.
The next node reads it
The Fix node takes triage.json as its input. The file stays in the library, with the run that wrote it in its history.
Every triage reads the same, whether Claude Code or Codex wrote it, so the next node and you read it without guessing.
Questions about the library
What formats can my context files for AI be in?
Can I define my own record types?
How do my agents get the right files?
Do I need a separate rule file for each coding agent?
Can I take my files with me?
Give your agents one library to work from
Bring your files, define the record types you need, and connect the coding agent you already use.