Features

What Veyyon includes.

TypeScript and Bun run the agent loop and TUI. Rust handles grep, PTY, and hashline edits. Here's what comes with it.

Editing

Editing.

Hashline anchors edits to file content instead of line numbers, and checks them before applying. Weaker models lose fewer turns to malformed diffs.

Content-anchored, natively verified edits with fewer retry loops on malformed diffs.

Ground, propose, execute — tools gated to non-mutating work until you approve the plan.

A review pass over the agent's own changes before they land, on a model you choose.

Control

Approvals and the OS sandbox are two separate layers — one asks before an action, the other enforces limits at the system level.

Read-only, workspace-write, or full access; per-tool approval tiers; fail-closed by default.

Landlock, seccomp, and Seatbelt back the workspace boundary — not just a prompt you can click through.

One-shot runs for CI and scripts, with the same approval model and honest exit codes.

Context

Memory, compaction, and session trees keep long sessions coherent across turns and across days.

Durable project memory across sessions — recall in, retain out. Off until you enable it.

Fork, clone, and label threads with /tree. Explore without losing the trunk.

Live multi-agent view — a model per subagent, drill in, watch the work as it happens.

Extend

Open interfaces for the rest of your stack — MCP servers, authored skills, lifecycle hooks, and a real browser.

Connect MCP servers, author SKILL.md skills, and run lifecycle hooks around every turn.

A real headless browser and multi-provider web search, in the loop when the task needs the web.

A self-test that reports what's missing, and Vim-mode keybindings you can remap per profile.

The thesis

Each feature here is about getting more real work out of the model you're running. The handbook covers how each one works.