Code, debug, and ship
from your terminal
An autonomous, self-learning AI coding agent that handles complex, multi-step development tasks — directly from your terminal, with the model of your choice.
╭────────────────────────────────────────────────────────────────────────────╮ │ >_ YottaCode by YottaDynamics (v0.3.1 · e3bbfd3) │ │ │ │ model nemotron-3-ultra-550b-a55b │ │ provider nvidia-nim │ │ directory ~/go/src/github.com/yottadynamics/yottacode │ │ context branch=main · mem=USER+YOTTA+UMEM+PMEM │ │ │ │ tip: /model <name> switches model on the fly — auto-switches provider when │ │ the name belongs to a different profile. │ ╰────────────────────────────────────────────────────────────────────────────╯ ╭────────────────────────────────────────────────────────────────────────────╮ │ ❯ ask anything… / commands · @ files · ↑↓ history │ ╰────────────────────────────────────────────────────────────────────────────╯ ● nemotron-3-ultra-550b-a55b · nvidia-nim · ctx ░░░░░░ 9.3K / 262K (4%)
Quick start
Install on Linux or macOS.
curl -fsSL https://yottacode.ai/cli/install.sh | bashRun the setup using the interactive wizard:
yottacode setupMemory that learns you
Plain-text, grep-able memory: you own USER.md, the agent curates the rest — project writes are approval-gated. Per-turn retrieval surfaces only what’s relevant, so it gets sharper the more you use it.
Use any model, switch any time
Native adapters for Anthropic, OpenAI, Gemini, xAI, and local Ollama, sign-in with your ChatGPT or GitHub Copilot subscription, or any OpenAI-compatible gateway. Swap mid-session — no lock-in, no hidden default that silently bills you.
A built-in security policy
First-launch trust prompt per workspace, approval modals with syntax-highlighted diffs, allow / ask / deny project rules, and write-path validation that firewalls secret-bearing paths like .env and ~/.ssh.
Deep GitHub integration
A typed GitHub adapter — no gh CLI required. Read, open, update, and review PRs, triage issues, and take an issue end-to-end: research → plan → branch → implement → tests → draft PR.
Plan first, then let it run
Plan mode investigates read-only and drafts a plan you approve; approving drops into auto mode that skips per-tool prompts while bash and commits stay gated. The agent can never escalate its own permissions.
A deep, repo-aware tool surface
50 built-in tools spanning reads, writes, search, a full git suite with risk-tiered approvals, bash, tests, and working-plan tracking — each with an explicit approval policy.
Rewind and resume
Every prompt auto-checkpoints the conversation and pre-edit files — a double-tap of Esc rolls back either or both. Sessions save atomically each turn, and /recall full-text searches all of them.
Delegate to typed subagents
Hand research, code search, planning, and verification to subagents that run in their own context window — the parent sees only the final answer. Add your own under .yottacode/agents/.
Reusable skills, on demand
17 built-in playbooks — TDD, security audit, performance profiling, remote ops, and more — load only when relevant. Install your own from a path, URL, or GitHub shorthand.
Work in parallel
Run each session in its own git worktree so two agents can edit the same repo without colliding; a per-repo .worktreeinclude carries gitignored configs into each one.
Make it your own
Drop a markdown file into .yottacode/commands/ to get /your-command with argument substitution and file references — and connect MCP servers for tools we didn’t ship.
Built for the terminal
Inline rendering keeps your scrollback intact — markdown output, a Tab-completing slash palette, image paste. Script the same core in one-shot mode, all from a single static Go binary for Linux and macOS.
Bring your own model — works with
…and any OpenAI-compatible API