Your models
Use OpenAI, Anthropic, Gemini, xAI, Ollama, OpenRouter, ChatGPT OAuth, or any OpenAI-compatible endpoint — and switch providers as your task, budget, or security posture changes.
Product · Open Source
yottacode
Sovereign AI coding from your terminal — local-first, on-prem capable, and cloud-ready.
yottacode is an open-source coding agent for engineers and teams who want AI acceleration without giving up control of their source code, models, infrastructure, or execution environment.
curl -fsSL https://raw.githubusercontent.com/yottadynamics/yottacode/main/install.sh | bash
Also available via go install. Release artifacts ship with verified checksums.
License MITLatest Release on GitHubPlatforms Linux · macOSRuntime Single Go binary
yottacode exists to give engineering teams sovereign AI coding: acceleration without surrendering control of source code, models, tools, permissions, memory, or deployment environment. Run it locally, on-prem, inside private networks, or through planned hosted cloud agents when that is the right operational fit.
Sovereign by design.
AI should adapt to your engineering environment — not force your engineering environment into one deployment model. yottacode is built around practical control surfaces that matter in real development work:
Your deployment
Run locally from a terminal-native binary today, connect to private model gateways where needed, and keep a path open for planned hosted cloud agents.
Your rules
Approval modes, write-path validation, permission policies, plan mode, and visible diffs keep execution understandable and controlled.
Your network
Support local, on-prem, private-network, and hosted-agent workflows without forcing every team into the same operating model.
Your context
Session recall, project memory, user memory, and agent-managed notes help durable knowledge compound while keeping context inspectable.
Production-minded capabilities, available now.
Multi-provider support
Native adapters for major AI providers, local Ollama models, ChatGPT OAuth, OpenRouter, and OpenAI-compatible endpoints.
Repository-aware tools
Read, edit, search, run tests, inspect diffs, manage git, and work through real codebases from the terminal.
Approval and permission modes
Normal, auto, plan, and YOLO workflows let developers choose how much control stays in the loop.
Planning and subagents
Use plan mode for design-first work, or delegate focused research, review, test, documentation, and implementation tasks to specialized subagents.
GitHub workflow
Review pull requests, inspect checks, open or update PRs, and work from issues through typed API-backed tools.
Memory and recall
Combine USER.md, project context, agent-managed memories, and full-text session search for durable, inspectable context.
Security-conscious execution
Write-path validation keeps edits inside the working tree, while approval policies help protect sensitive files and high-risk actions.
Extensibility
MCP client support, built-in skills, and custom slash commands make repeatable team workflows possible.
Parallel work
Worktree-backed sessions and experimental dispatch fan-out let multiple agents work on isolated branches without colliding.
Install options
One-line install, go install, and verified release checksums.
Try yottacode on a real repository.
Install it, point it at your codebase, and tell us what to improve. yottacode is built in the open by YottaDynamics — a small team making the terminal the best place to work with AI.