Overview

Venna Code is the agentic Code mode in the Venna desktop app. It pairs a model with a curated toolset — file I/O, bash, glob, grep — and runs a request → tool-call → re-prompt loop: the model proposes a tool call, you approve it (or it auto-approves), the result feeds back, and the loop repeats until the task is done.

How a turn works

• Auto vs. approved. Read-only tools (glob, grep, file reads) auto-approve; anything that writes to disk or spawns a process waits for your approval. Every call streams to a live transcript.
• Permission modes. plan (read-only until you approve a plan), default (approve writes as they land), or acceptEdits (auto-approve edits).
• Sub-agents. The main agent can spawn one sub-agent to take a focused task and report back.
• Sessions. Each session is journaled as append-only JSONL under ~/.venna/code/sessions/ — a durable, replayable transcript.

Extend Code mode

Add capabilities through four directories under ~/.venna/: skills, plugins, hooks, and MCP servers.