Your engine.
Your models.
Your rules.
A portable engine for AI agents. Connect a model, give it a workspace, and shape how it works. One executable. Three ways to run.
$ zagent --session weekly -p
"Summarize the reports in this folder."
session weekly · your configured model
↳ list_dir path: ./reports
↳ view_file path: report.csv
↳ Preparing a summary from the supplied data
✓ Draft ready for your review.
Session saved. Continue whenever you need.
$ zagent --session weekly --ui cli
Attach to a local engine from your terminal.
you Add a short executive summary.
agent Preparing a revised draft…
/note Use British spelling.
/status Show context and running tasks.
/exit Detach from the interface.
$ zagent --session weekly --ui web
Opens an authenticated browser UI on loopback.
View the conversation and tool results.
Add supported attachments.
Send notes as the agent works.
Same session. Same engine.
Closing the page detaches the interface.
Illustrative workflow · output depends on your model and tools
Choose your
model connection
The workspace is yours.
So is the way you work.
Run a task from a script, work alongside it in the terminal, or follow the conversation in your browser.
A local engine. A shared session. Your configuration.
Pick up the conversation.
Named sessions retain history and attachments. Detach an interface and return to the local engine later.
Keep context under control.
Configure attachment pruning, request-size limits, and compaction. Inspect context and usage as you work.
Add the capabilities you need.
Enable a small tool set, connect your own programs, and use background tasks or subagents for bounded work.
Make it your agent.
Without rebuilding it.
Set the behavior in Markdown. Configure the engine in JSON. Extend its tools in the language you already use.
# Your campaign reporting assistant Prepare draft reports from workspace files. Use the supplied data. State assumptions. Keep source references with each finding. Ask for review before external publication. # A skill, ready when the task calls for it skills/ campaign-report/ SKILL.md
Small setup. Open possibilities.
Download your build.
Choose your OS and processor. Verify the checksum and unpack the executable.
Windows · Linux · macOS
x64 · ARM64Bring your model.
Initialize the home, add your endpoint and model, then test the connection.
zagent init
zagent doctorGive it a workspace.
Start with non-sensitive files and a narrow tool set. Review the result of your first task.
zagent --session demo
-p "Summarize this folder."One download.
Your platform.
v0.1.1 · Released 15 October 2026
Self-contained engine. No Go installation required.
A few things, before you run.
Is the engine source included?
No. Downloads contain the compiled executable, quick-start information, and third-party notices. You can customize the documented configuration and extension interfaces without engine source.
Do I need an AI provider account?
You need access to a supported model endpoint. Bring your own provider credentials, use a gateway, or connect a compatible local server. Model-provider charges are separate from the engine.
Can it run unattended for my clients?
This development preview is intended for supervised evaluation. Permissions, completion checks, isolation, and other operational controls need work before unattended client deployment. Read the security and agency workflow guides.
Where does my data go?
The engine runs locally. Prompts, tool results, and enabled attachments are sent to your configured model endpoint. External tools may contact other services. Keep confidential data out of preview evaluations.
Can I add tools without changing the engine?
Yes. Add a JSON manifest and a program under the home tools directory. Use Python, Node, PowerShell, Bash, or a compiled program. Any required tool runtime must be installed separately.
Build your way of working.
Start with a model, a workspace, and a well-defined task.