Your AI agent, the same in every environment: its workflow, tools and access come with it, and its keys never reach it. That is where it is going. Today it shrinks noisy tool output, folds what an agent already read, and redacts credentials in the output it rewrites. Nothing is lost.
🇺🇸 English | 🇯🇵 日本語 | 🇨🇳 简体中文 | 🇸🇦 العربية | 🇮🇩 Bahasa Indonesia | 🇻🇳 Tiếng Việt | 🇰🇷 한국어
brew install fajarhide/tap/contexere && contexere init
OMNI is now Contexere, Latin for "to weave together" and the root of "context". The
omnicommand keeps working until 0.9.0; runcontexere init --<host>for each agent you use to move an existing install across. Since 0.8.6 your data moves to~/.contexereon its own.
The goal: take Contexere to a new VPS or laptop, and your agent arrives knowing its workflow, with its access to MCP servers and APIs working. The keys behind that access are never shown to the agent, so they never leak. Around that sit skills you set up once for every agent, and memory that carries across agents and machines. That part is on the way. What ships today is the first piece, the context and token layer described below.
A coding agent reads everything a command prints: progress bars, layer hashes, a file it already read ten minutes ago. Every byte is billed and takes room the agent needs for the actual work.
Contexere takes out what the model does not need, before the model reads it. Nothing is deleted: whatever it removes stays on your machine, and one command brings it back.
- Reads a file the agent already read: sends one line with a handle instead of the file again.
- Runs a noisy build or test: keeps the result and the failures, drops the progress noise.
- Prints a token or password in output it rewrites: replaces it with
[REDACTED]before the model sees it. - Fails a command: passes it through as printed, so the real error reaches the agent. A credential printed by a failing command reaches the agent too.
- Reads JSON, YAML or CSV: leaves it byte for byte.
contexere doctorchecks that each agent's hooks fire and redact.
It costs about 21 ms per command. Savings depend on what your agent does; run contexere stats for your own figure. Method and full numbers are on the benchmarks page.
brew install fajarhide/tap/contexere| Where | Command |
|---|---|
| macOS, Linux, WSL | curl -fsSL omni.weekndlabs.com/install | bash |
| Windows (PowerShell) | irm omni.weekndlabs.com/install.ps1 | iex |
Existing installs keep working under omni, the same program, until 0.9.0. Run contexere init --<host> for each agent to move them across (what changes).
Inside Claude Code, /plugin marketplace add fajarhide/contexere then /plugin install contexere@omni adds a skill that installs and checks it for you.
contexere init # pick your agents: Claude Code, Codex, Cursor, VS Code and more
contexere doctor # check it is wired, or `contexere doctor --fix`
contexere stats # what Contexere saved on your own history
After init, the hooks run on their own. There is nothing to prefix.
| Tier | Agents |
|---|---|
| Full (rewrites tool output before the model reads it) | Claude Code, Codex CLI, Gemini CLI, OpenClaw, Hermes, Pi, Aider (via cmd | contexere) |
Handoff-first (shortens what the agent runs through contexere_run) |
Cursor, Windsurf |
MCP-only (memory and recall, plus contexere_run) |
Cline, Roo, OpenCode, VS Code, Zed, Copilot, Antigravity |
Codex runs only hooks it trusts: after contexere init --codex, start codex once and approve them (#359).
Issues and pull requests are welcome. Start with CONTRIBUTING.md.
Do not open a public issue. Report privately; details in SECURITY.md.