Experimental modules and skills for flex — the SQLite knowledge and memory engine for AI agents.
The contract, in one line: everything here works for us, installs over a stable flex, and may change or break without notice. Issues and field reports are the whole point. No support promise.
CORE ships in the flex wheel — actively developed
BETA ships in the wheel — CI-tested, issue-driven fixes
LABS this repo — install over the wheel, experimental
Labs is where things earn their way into the wheel. The graduation bar is real-world execution: a labs module that accumulates field reports and passes the cold-start test matrix moves to the community tier in a flex release.
Requires flex >= 0.40 (curl -sSL https://getflex.dev/install.sh | bash).
flex discovers external modules in ~/.flex/modules/. Installing a labs
module is a copy:
git clone https://github.com/damiandelmas/flex-labs
cd flex-labs
./scripts/install-module.sh devto # copies modules/devto -> ~/.flex/modules/devto
flex init --module devto --tags python,ai
./scripts/install-module.sh aider # agent modules resolve from modules/agents/
flex init --module aiderSkills install the same way to your agent's skills directory:
cp -r skills/flex-topology ~/.claude/skills/
cp -r skills/flex-archeology ~/.claude/skills/
cp -r skills/flex-digest ~/.claude/skills/| module | what it indexes | one-liner |
|---|---|---|
devto |
Dev.to articles + comments by tag/author (no auth) | flex init --module devto --tags python,ai |
lobsters |
lobste.rs stories + comments by tag (no auth) | flex init --module lobsters --lobsters-tags programming,ai |
agents/aider |
Aider chat histories — programmable memory for Aider | flex init --module aider |
agents/opencode |
OpenCode SQLite sessions — programmable memory for OpenCode | flex init --module opencode |
Every module cell self-documents: run @orient on it and the module's own
instructions, views, and presets come back as part of the answer. Known gaps
are documented per module — read them; they're honest.
| skill | what it does |
|---|---|
flex-topology |
standup-style map of your recent work — workstreams, hubs, what's hot, what went quiet — mined from your own coding-agent sessions |
flex-archeology |
project history reconstruction from session evidence: file lineage, workstream recovery, artifact genealogy |
flex-digest |
coworker-style standup of recent agent activity — inventory, file hotspots, delegations, open loops — four queries + synthesis |
All three expect claude_code and/or codex cells (the flex coding-agent install).
Open an issue with: the command you ran, what you expected, what happened, and
flex status output. A failed experiment reported well is a contribution.
MIT, same as flex core.