Codex alternatives for repository coding agents

OpenCode replaces Codex repository work with a model-flexible terminal or IDE agent. A planning agent explores code without editing by default, while the development agent changes files. Automatic language-server context and parallel sessions support work across a project.

Qwen Code suits Codex tasks that need planning, code changes and agent collaboration. It offers subagents, agent teams, MCP tools and isolated worktrees, with headless operation for scripts. Local models through Ollama or vLLM and several hosted providers are available.

Open Interpreter is a Codex-derived terminal agent with configurable models and provider-specific agent behavior. It supports the Codex exec protocol and repository instructions, with QA skills for checking web or native apps.

Codewhale suits Codex users who want repository edits with explicit permissions and recovery controls. Read code in planning mode, review diffs and command results, then resume a saved session from the terminal or local browser. Fleet divides larger tasks among agents with different roles and models.

Cline brings Codex-style repository tasks into an existing editor, a terminal or its desktop app. Plan and Act modes separate the proposed approach from execution, while diffs and checkpoints support review. The CLI also runs recurring work in scripts and CI, with local or hosted model connections.

Kilo Code covers Codex coding tasks in an IDE or CLI, with planning, debugging and coordinated file changes. Run parallel work in isolated worktrees and switch models during a task. It also offers inline completions when you want the agent alongside everyday editing.

OpenHands covers the delegated and recurring engineering side of Codex. Agents investigate failures, apply changes, run checks and prepare pull requests, with repository events or schedules starting work. Run its own agent on infrastructure you control; selecting Codex as the backend would keep Codex in the workflow.

cc-haha gives repository agents a desktop workspace with file-by-file diffs, whole-turn undo and an embedded browser for previews. Separate sessions and Git worktrees keep tasks apart; model connections include Ollama and LM Studio.

Goose suits Codex tasks that combine coding with research or other connected tools. Work from its desktop app or CLI, use MCP extensions and choose Ollama or a hosted provider. It is a broader personal agent rather than a dedicated cloud coding service.

Aider suits Codex users who prefer a terminal assistant closely tied to Git. A repository map supplies code context, edits become reviewable commits, and lint or test feedback can guide follow-up changes. Choose a local or hosted model for the task.

Prime Agent covers repository work with persistent Python sessions and cooperating subagents. Background tasks survive a terminal disconnect, and saved goals or schedules let work continue within limits you set.

MonkeyCode suits teams delegating repository tasks into managed development environments. Agents can build, test and preview changes, with automated pull-request review and mobile access to ongoing work.

Fast Agent covers terminal coding and scripted agent tasks with shell tools and your chosen model. You can connect the same agents to compatible editors through ACP or reuse them in Python workflows.

Pydantic DeepAgents covers repository changes with checkpoints and reviewable file edits. Parallel branches keep competing approaches separate, and an optional judge can use test results to help choose a continuation.