Everything you need.
Nothing you don't.
Core capabilities are built in from the OMP mainline; advanced content is ported from Pi's semantic source. Extensions cover the rest — build what you want, install what you don't want to build.
Core capabilities
SSubagents
Split work across parallel subagents, each with its own context and toolset. Results aggregate back into the main session.
PPlan mode
Plan before you act: the agent produces a plan for your review, then touches the workspace only after approval.
LLSP language services
Built-in LSP client: go-to-definition, find references, diagnostics, and completions — code semantics at edit time.
DDAP debug adapter
Speaks the Debug Adapter Protocol: set breakpoints, read stack frames, inspect variables, and explain failures.
HHindsight memory
Lessons and outcomes from past sessions are revisited and applied automatically in later tasks.
EHashline edits
Line-precise, content-addressed editing semantics: refactors and bulk replacements stay traceable.
TTime-traveling rules
Project rules that apply per time slice: old commit conventions don't leak onto new code.
RNative Rust engine
Text, regex, grep, and diff heavy lifting on Rust primitives — large files and big repos stay fast.
TProject tracking docs
A long-term tracking system independent of memory: activate with /tracking, the AI writes milestones and key decisions into .zeta/tracking/, survives across sessions, visualized in the desktop Tracking panel.
ZZH/EN UI switch
/language zh toggles the CLI and command-description language; the setting persists.
WDesktop default workspace
The desktop app ships a global default workspace (~/.zeta/workspace) — ready to use without picking a directory first.
CClear config paths
Project-level .zeta/ and user-level ~/.zeta/ (same location on Linux and Windows — under the home directory): config, memory, tracking, and logs each in their place.
Context engineering
AGENTS.md
Project instructions loaded at startup from ~/.zeta/agent/, parent directories, and the current directory.
SYSTEM.md
Replace or append the default system prompt per project — a persona and constraints for each repository.
Skills
Capability packages loaded on demand: instructions plus tools, with progressive disclosure that doesn't bust prompt caches.
Prompt templates
Reusable Markdown prompts, expanded with /name.
Auto compaction
Older messages are summarized near the context limit; strategy is fully customizable via extensions.
Dynamic context
Extensions inject messages before each turn, filter history, implement RAG, or build long-term memory.
Sessions: a tree, not a log
Branching history
Sessions are stored as trees. Navigate to any earlier point and continue from there.
Export & share
/export produces HTML; /share uploads to a shareable URL anyone can open.
Encrypted collab
/collab opens an end-to-end encrypted live session — participants render the same session and tool cards.
Resume anywhere
Restart with --continue to re-enter the same session with full reasoning and decisions intact.
Four modes
The full TUI: panels, previews, tool cards, keybindings.
zeta -p "query" for scripts; --mode json streams events for consumers.
JSON protocol over stdin/stdout for non-Node integrations.
Embed as a library in your own application, with full control over the agent lifecycle.
Primitives, not features
Permission gates
Custom confirmation flows — allow or block sensitive operations per your environment and security needs.
Path protection
Declare protected paths; the agent's reads and writes stay inside the allowed region.
MCP integration
Connect MCP servers via extensions, or build CLI tools with READMEs as skills.
Sandboxed execution
Run commands inside a container or sandbox, isolating risk from the agent.
Extension discovery
skills, hooks, tools, commands, rules, prompts, and .mcp.json all discovered from extensions.
50+ examples
Subagents, plan mode, SSH execution, custom editors, status bars, overlays — official examples ready to use.