Bring your model
Use Ollama locally, authenticated coding CLIs in headless mode, OpenAI-compatible endpoints, or your own process adapter.
█████╗ ██╗ ███╗ ██╗ ███████╗ ██╔══██╗ ██║ ████╗ ██║ ╚══███╔╝ ███████║ ██║ ██╔██╗ ██║ ███╔╝ ██╔══██║ ██║ ██║╚██╗██║ ███╔╝ ██║ ██║ ██║ ██║ ╚████║ ███████╗ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═══╝ ╚══════╝
A local-first agent harness and session multiplexer. Run Ollama, LiteLLM, Codex, Claude Code, custom endpoints, tools, plugins, and subagents from one terminal-native workspace — with memory that survives the session.
brew install wess/packages/ainzInspired by irssi and BitchX: messages share a line with their nick, agents live in the right-hand roster, and the channel topic says what the selected agent is doing. Switch the rail to the machine-wide Synapse mesh with Cmd+R. Markdown renders in the terminal, and transcript text remains selectable with the terminal's normal copy command.
#main · Improve the output
15:31 <you> Make the buffer easier to read.
15:31 <Ainz> Keep names beside their messages. Longer replies
wrap under the text, with room for code and lists.
15:32 ▸ shell cargo test*** Completed · 8s
Working, Responding, Running tools, and Waiting for approval tell you when a run is active. Completed, Cancelled, or Failed stays above the prompt after it ends, with a matching marker in the buffer. Completion includes end hooks.
Enable Automatic improvement in settings or use /improve on once. Normal saved runs then trigger background experiments that propose harness fixes and reusable skills. Fixed behavioral tasks check each candidate before you review and accept it.
Set up automatic improvement and learn how source checks, behavioral evaluations, and review protect the next run.
Choose /header mascot for the built-in mascot. Draw your own artwork or tune the interface colors in the designers, then copy one install command and select your creation inside Ainz.
Artwork guide · Theme guide · Buffer and completion states
Use Ollama locally, authenticated coding CLIs in headless mode, OpenAI-compatible endpoints, or your own process adapter.
Resume branching sessions, preserve token totals, compact long histories, steer a run at safe turn boundaries, and export any branch to Markdown.
Durable memory recalled when a session opens, search across earlier sessions, and skills a session writes for the next one — kept locally, or in Synapse.
Delegate durable subagent runs, leave them working in the background, and switch among their transcripts without leaving the main terminal.
Import the tool servers your other tools already have, then add skills, prompts, and capability-scoped plugins with content-pinned approval.
Interactive TUI, one-shot prompts, JSON streams, JSON-RPC, or your own Rust host. The core runs without the terminal and lets you choose which plugin runtimes to include. Embedding guide.
Walk earlier prompts with Up and Down, edit by word or line, and add a newline with Shift+Enter. @ completes a path, and Esc Esc rewinds to the last one.
A permission prompt shows a diff for an edit, or the command for a shell call, and takes y, n, or a to keep the decision for the rest of the run.
A command can run at a session's start, before or after any tool call, and when it ends — a pre_tool hook can block the call outright.
ainz synapse connect links both products in one step. Synapse shares Ainz's memory, guidance, and skills with your other tools; --mesh gives every subagent its own visible seat.
ainz. The setup screen lists providers and pulls their models straight from each endpoint — no config file, no typing a model name by hand.