AI coding agents for repository tasks

Terminal, cloud and dedicated desktop workflows for reviewable code changes

Updated Nov 2026

These products take a task on an existing repository, edit files, run commands and return the changes for review. Entries name the workflow and who makes each product.

Choosing a workflow

Terminal agents work with a local repository and its development tools. Cloud agents can continue tasks in a separate environment. Some products provide both. The entries identify the qualifying workflow and link to current documentation. Review the execution environment, repository permissions and resulting changes when evaluating a product.

Products in alphabetical order

Aider Aider

Aider pairs a terminal session with a language model to change an existing codebase or start a project. It can connect to cloud or local models and map the repository for context.

Its Git integration commits changes and lets developers inspect or undo them with Git tools. Aider can also run linting and tests after edits and work on the failures those checks find. The source code is public.

[Aider is free and open source](https://aider.chat/). Model API usage and any local infrastructure costs are separate. · open-source · terminal · git

Amp Amp

Amp, from Amp Frontier Corporation, is a coding agent and development environment. It creates an orb for each thread, with code and tools set up, so work can continue after the developer closes the laptop.

Developers can start in the web app, follow up from mobile or desktop apps, and use the CLI on their own machine. Threads stay available across those interfaces for continued work.

[Megawatt is USD 20 per month](https://ampcode.com/pricing), or USD 10 per month for eligible students and teachers, and Gigawatt is USD 200 per month. Hobby is free with your own runners or self-hosted orbs, while Amp-hosted orbs have usage charges. Team workspaces have no extra fee, Enterprise pricing is on request, and external model subscriptions or keys retain their provider charges.

Augment Code Augment Code

Harness has acquired selected Augment Code assets, including Cosmos, Auggie CLI and the Code Context Engine. Cosmos is becoming the Harness Cosmos Software Factory Agent, which coordinates planning, implementation, tests and pull requests for a team's merge decision.

Auggie CLI remains available for interactive repository work and scripted automation. It uses codebase context, supports repository rules and tool permissions, and can connect engineering tools through MCP.

For teams buying around a ticket-to-pull-request workflow, Cosmos can start from ticket updates, pull requests, schedules or webhooks and continue through review feedback, failed checks and merge conflicts. The intended handoff is a change prepared for the team’s merge decision.

[Standard is USD 20 per month](https://www.augmentcode.com/pricing), and Business is USD 100 per month, each for up to 50 seats with no per-seat fee. They include USD 20 and USD 100 of monthly usage respectively, with pay-as-you-go top-ups. Enterprise pricing is on request.

BLACKBOX AI BLACKBOX AI

BLACKBOX AI's CLI works directly in a codebase. Its guides cover debugging, applying fixes, running tests, refactoring and Git worktrees. Remote agent sandboxes run the same kind of work in the cloud.

The company sells annual enterprise commitments billed per token across the CLI and cloud agents.

Claude Code Claude Code

Anthropic's Claude Code reads codebases, edits files and runs commands. It works in the terminal, supported IDEs, the desktop app and the browser, with permission rules that control tool use.

For repository maintenance, it can write tests, fix lint errors, resolve merge conflicts and update dependencies. Developers can review plans and diffs in supported interfaces and use its Git workflows to prepare commits or pull requests.

[Pro is USD 20 per month or USD 200 per year](https://claude.com/pricing), Max is USD 100 per month, and Ultra is USD 200 per month; Free excludes Claude Code. Team Standard seats are USD 25 per month or USD 20 per month billed annually, and Premium seats are USD 125 per month or USD 100 per month billed annually. Enterprise is USD 20 per seat per month billed annually, with a 20-seat minimum and usage charged at API rates. · terminal · anthropic · agent

Cline Cline

Cline, from Cline Bot Inc., is an open source coding agent for editors, the terminal and desktop. It reads project context, makes coordinated file edits and runs shell commands, including tests. Developers can approve each step or configure automatic approval, and choose hosted models, compatible endpoints or local models.

[Cline is free for individual developers](https://cline.bot/pricing). AI inference is billed by usage through Cline credits or your own provider keys, with no subscription or seat fee for the open source version. · open-source · vscode · byo-model

CodeBuddy Code CodeBuddy Code

Tencent Cloud released CodeBuddy Code, a terminal agent, on 9 September 2025. Its commands cover plans, code changes, diff review, permissions and test and build checks. Pro costs $10 a month, with credits shared between the IDE and the CLI.

Cosine Cosine

Cosine offers CLI, desktop and cloud agents for codebase work, including writing software, refactoring and running tests. Its wider product range includes models for organisations that need deployment control.

For delegated repository tasks, Cosine Cloud runs agents on isolated workers or a team's own SSH targets. Diffs appear for review, and commits, pushes and pull requests require approval. Teams can select the model and scope for each task.

[Starter starts at USD 19 per month](https://cosine.sh/pricing), Team is USD 199 per month, and Enterprise is USD 999 per month. Their monthly allowances are 4 million, 47 million and 240 million credits respectively, with add-ons at USD 6.50, USD 5 and USD 4.50 per million credits. Private enterprise deployments are priced on request.

Crush Crush

Crush, from Charm, is a terminal app that views and edits files, runs Bash commands and keeps project context, with approval before tools run. Users bring their own model provider or use Charm's Hyper subscription. The source is published under the FSL-1.1-MIT license.

Cursor Cursor

Cursor, made by Anysphere and acquired by SpaceX, has desktop, CLI and cloud agent workflows. Its cloud agents use their own computers to build, test and demonstrate changes for review.

In August, OpenAI said it would stop supplying its models to Cursor after the SpaceX deal.

Hobby is free with limits. Pro is USD 20 per month, Pro+ USD 60 and Ultra USD 200; Teams is USD 40 per user per month. Further model usage can cost extra. [source](https://cursor.com/pricing). · ai-editor · agent · hybrid

Devin Devin

Devin, from Cognition, takes delegated development tasks through Cloud, Desktop and CLI. Its development environment has a shell, code editor and browser, so it can work through implementation and testing with project tools.

Teams can delegate work to the cloud and iterate locally, with the resulting code available for review.

[Pro is USD 20 per month](https://devin.ai/pricing), Max is USD 200 per month, and Teams is USD 80 per month plus USD 40 per full developer seat per month. Free includes limited agent use, unlimited inline edits and Tab completions; Enterprise pricing is on request. Extra paid usage is billed at API rates. · autonomous · cloud · enterprise

Factory Droid Factory Droid

Factory's Droid plans, builds and validates code changes. Its CLI works beside Git, tests and the editor, with project context, approvals and MCP tools in an interactive terminal interface.

Teams can review plans, tool calls and diffs, then use headless execution for automation. Factory also offers desktop and web surfaces, deployment control and model choice.

[Pro is USD 20 per month](https://factory.com/pricing), Plus is USD 100 per month, and Max is USD 200 per month. Teams is USD 60 per team per month plus USD 40 per seat per month. Business and Enterprise pricing is on request; paid plans have usage limits. · enterprise · autonomous · droids

Gemini CLI Gemini CLI

Google's Gemini CLI is an open source terminal agent that can query and edit codebases, debug issues and run non-interactive tasks in scripts. Its MCP support connects additional tools. Developers can use the same terminal workflow for repository exploration and repeatable automation, with the project maintained in Google's public GitHub organisation.

Since 18 June 2026, free users and Google AI Pro and Ultra subscribers can no longer use it; Google points them to Antigravity CLI. Paid Gemini API keys and Gemini Code Assist Standard and Enterprise still work.

[Gemini Code Assist Standard is USD 22.80 per user per month](https://cloud.google.com/products/gemini/pricing?hl=en) with a monthly commitment or USD 19 with a 12-month commitment; Enterprise is USD 54 or USD 45 per user per month respectively. The Gemini CLI application is open source and free, but its former free individual model access has moved to Antigravity CLI. Legacy API and Vertex AI access has separate usage charges.

GitHub Copilot GitHub Copilot

GitHub Copilot can take an issue through implementation to a pull request, with code review and follow-up changes in the same repository workflow. Developers can start a session with the issue discussion and repository context already attached.

Its sessions continue across the app, CLI, editor, web and mobile. Parallel sessions let teams track separate tasks, while local and pull request reviews provide places to inspect the resulting changes.

[Pro is USD 10 per user per month](https://github.com/features/copilot/plans), Pro+ is USD 39 per user per month, and Max is USD 100 per user per month. Free includes 2,000 completions per month and limited model access. Paid plans include usage credits, with additional GitHub AI Credits at USD 0.01 each. · ide-plugin · agent · incumbent

GitLab Duo Agent Platform GitLab Duo Agent Platform

GitLab Duo Agent Platform has flows that convert issues into merge requests and fix failed CI/CD pipelines. It also provides agentic file editing, code review and custom agents inside GitLab's development workflow.

Access depends on GitLab version, enabled settings and project permissions. Self-managed deployments need instance configuration, and local use requires an authenticated editor extension and an appropriate project role.

[GitLab Credits cost USD 1 each](https://about.gitlab.com/pricing/). GitLab Free is USD 0, Premium is USD 29 per user per month billed annually, and Ultimate pricing is on request. Premium and Ultimate currently include promotional monthly allowances of 12 and 24 credits per user respectively; Duo Agent Platform usage consumes credits.

Google Antigravity Google Antigravity

Google Antigravity has an app, CLI and IDE for agent-based development. The app groups conversations into projects and manages parallel local agents across workspaces. Its CLI runs coding agents, shell commands and background subagents.

Developers who want to inspect the work in an editor can use the IDE's agent manager and artifacts. An SDK is available for custom agents and engineering automation.

[Individuals is free at USD 0 per month](https://antigravity.google/pricing), with basic weekly limits. [Google AI Pro is USD 19.99 per month](https://gemini.google/us/subscriptions/?hl=en), and Google AI Ultra is USD 99.99 or USD 199.99 per month for higher limits. Business access through eligible Gemini Enterprise Standard/Plus plans starts at USD 30 per seat per month, or has no seat fee with consumption-based billing. · agent-first · google · new

Google Jules Google Jules

Google's Jules clones a selected GitHub repository into a cloud VM and develops an implementation plan. Give it a detailed prompt, or assign work with the Jules label on a GitHub issue. It returns a diff and creates a pull request so developers can inspect the edits before merging.

[Jules is free](https://jules.google/) for up to 15 tasks per day and three concurrent tasks. Jules in Pro and Jules in Ultra raise those limits through Google AI subscriptions: [Google AI Pro is USD 19.99 per month](https://gemini.google/us/subscriptions/?hl=en), and Google AI Ultra is USD 99.99 or USD 199.99 per month. · async · github · google

Goose Goose

Goose is an open source, local agent with desktop, CLI and API interfaces. It supports multiple model providers and MCP extensions for connecting tools.

Block developed and contributed Goose to the Agentic AI Foundation. Teams evaluating repository agents can inspect its source and configure the models and extensions used in their local workflows.

[Goose is free and open source](https://github.com/aaif-goose/goose). Model-provider usage, model subscriptions and local infrastructure costs are separate.

Grok Build Grok Build

Grok Build is SpaceXSI's open source coding agent and terminal interface. Its tools read, search and edit code and run commands, with skills, plugins, hooks, MCP servers and subagents available in the harness.

Its plan mode blocks edits until the developer approves the plan, and approved changes appear as diffs. Teams can also compile the harness and point it at local inference.

Free access is offered with limits. SuperGrok is USD 30 per month and SuperGrok Plus USD 100, with higher usage across the product family. [source](https://x.ai/pricing).

IBM Bob IBM Bob

IBM Bob includes Bob Shell, a terminal agent that reads and edits repository files, runs shell commands and shows proposed edits as a diff. Custom modes set which tools it may use. IBM aims it at fixes, refactoring and modernization of older code, and also sells an IDE version, managed SaaS and self-hosted deployment.

IBM announced general availability on 28 April 2026. Pro costs $20 a month with 50 Bobcoins, and a 30-day trial is available.

JetBrains Junie JetBrains Junie

JetBrains Junie runs repository tasks from the terminal, IDE workflows, GitHub Actions and GitLab CI/CD. Its plan mode writes requirements, design and delivery stages before implementation, with plans stored as editable project files. Developers can approve or redirect the plan and steer an active task, and choose supported models with their own keys.

[Individual AI Pro is USD 8.33 per user per month billed annually](https://junie.jetbrains.com/), and AI Ultimate is USD 25 per user per month billed annually. Junie Lite is free, and bring-your-own-key use has provider charges with no markup. Paid plans include AI credits and allow top-ups. · jetbrains · agent · ide-native

Kilo Code Kilo Code

Kilo Code is owned by Anaconda. Its open source agent runs in IDEs, a CLI and the cloud, with model choice and support for developers' own keys.

The current range also includes a desktop app in beta. Teams can choose the interface that fits their repository work while using the same agent across editor sessions, shell workflows and browser-started tasks.

[Teams is USD 15 per user per month](https://kilo.ai/pricing), optional Kilo Pass inference subscriptions start at USD 19 per month, and Enterprise pricing is on request. The open source application is free, with free or local models and your own provider keys supported. Hosted inference and cloud compute are billed separately, and credit purchases have a 5% processing fee. · open-source · vscode · agent

Kimi Code Kimi Code

Moonshot AI's Kimi Code reads a project, changes source files and runs shell commands, builds and tests. The CLI was rewritten in TypeScript and first released in May 2026; a desktop app that shows tool activity, changes and approvals followed on 17 September.

In China, Code access starts with the Plus membership at ¥99 a month.

Kiro Kiro

Kiro, from Amazon Web Services, turns requirements into specifications, designs and implementation tasks. Specs work in its IDE, CLI and web surfaces, with task status tracked in project files. Its agents can plan, build, review and verify work locally or in the cloud.

Teams can configure hooks to run linters, formatters and checks after file changes, or validate code before a task finishes.

[Kiro Pro is USD 20 per user per month](https://kiro.dev/pricing/), Pro+ is USD 40, Pro Max is USD 100, and Power is USD 200 per user per month. Kiro Free includes 50 credits per month. Team credit pooling is USD 20 per seat per month for at least two seats, with 500 credits per seat; paid-plan add-on credits cost USD 0.04 each.

Meta Muse Code Meta Muse Code

Meta's Muse Code runs in the terminal inside an existing project. It plans a task, edits files and runs commands, with approvals and a sandbox, and a headless mode for scripted tasks.

Meta released the public beta on 5 August 2026. Plans start at $5 a month.

Mistral Vibe for Code Mistral Vibe for Code

Mistral Vibe for code builds multi-file changes and generates tests. Developers can use its CLI, supported editors through ACP, or remote coding tasks in Vibe. Its asynchronous agents write code, test it and open pull requests, and remote work can return to the CLI for follow-up.

Teams can connect repository and project context from GitHub, GitLab and Jira.

[Pro is USD 14.99 per month](https://mistral.ai/pricing/), Team is USD 24.99 per user per month, and Enterprise pricing is on request, excluding applicable taxes. Free includes limited coding sessions. Eligible verified students can get Pro for USD 5.99 per month for up to 12 months.

Ona Ona

Ona, formerly Gitpod, is now part of OpenAI. It runs background agents in full cloud development environments: a team assigns a task on an existing repository and gets a pull request back to review. Environments run in Ona Cloud or in the customer's own VPC. Pricing goes through sales.

OpenAI Codex OpenAI Codex

OpenAI's Codex CLI inspects repositories, edits files and runs local development tools. Developers can use it interactively or run repeatable work through scripts and CI, with permissions configured for each run.

Codex also has ChatGPT desktop app, IDE and cloud workflows. From the terminal, developers can submit work to a configured cloud environment and apply the resulting changes to the local repository for review.

[Plus is USD 20 per month](https://learn.chatgpt.com/docs/pricing), Go is USD 8 per month, and Pro has USD 100, USD 200 and USD 500 monthly options. Free includes limited Codex access in the desktop app, subject to rollout. Business is USD 25 per user per month or USD 20 per user per month billed annually for at least two users; Enterprise and Edu pricing is on request, and API-key usage is billed separately. · cloud · terminal · openai

OpenCode OpenCode

OpenCode is an open source coding agent with terminal, desktop and web interfaces. It runs on the developer's machine and supports a choice of model providers. The desktop app has sessions, diffs and a built-in browser; developers can also steer and approve sessions from a phone.

[Go is USD 10 per month](https://opencode.ai/console/go) for access to selected hosted models. OpenCode is open source and free, and Console team controls are free. Other hosted inference is billed by model usage, while external model providers charge separately. · open-source · terminal · provider-agnostic

OpenHands OpenHands

OpenHands is an open source platform for software development agents. Agents plan work and apply changes across a codebase, and teams can run the platform locally, self-host it or use its cloud offering.

Its MIT-licensed core gives teams an inspectable foundation when deployment control is part of the buying decision.

[Teams is USD 500 per month](https://www.openhands.dev/pricing) for up to 10 users, plus USD 50 per additional seat per month, with a 30-day trial. Agent Canvas is open source and free to self-host with your own model key. Enterprise pricing is on request; model costs are separate. · open-source · self-host · sandbox

Poolside Poolside

Poolside's pool is a terminal coding agent that can use an OpenAI-compatible endpoint or a local model. It entered preview on 28 April 2026. The Desktop Assistant for macOS, launched on 28 July 2026, keeps repository sessions, commands, worktrees and diffs in one window.

Qoder Qoder

Alibaba's Qoder has a separate CLI for macOS, Linux and Windows that reads files, changes code and runs commands, from exploring a codebase to the commit. Permissions control sensitive actions. Cloud agents and a desktop app sit alongside it.

Lite costs $10 a month with 1,000 credits and Pro $20 with 2,000, shared across Qoder products.

Qwen Code Qwen Code

Qwen Code, released by Alibaba, is an open source coding agent with terminal, desktop and editor interfaces. It supports several model API protocols and can use third-party providers or local models.

For repository work, developers can ask it to explain a project, submit changes, review code and run tests. Its web interface is experimental, and scripts and SDKs provide routes for automation and custom integrations.

[Qwen Code is free and open source](https://github.com/QwenLM/qwen-code). Model access uses a paid Qwen Plus or Pro subscription, a provider API key or a local model; provider and infrastructure charges are separate.

Replicas Replicas

Replicas runs Claude Code, Codex and other coding agents in isolated cloud machines. Tasks start from its dashboard, Slack, Linear or a Git host. Each agent gets the repository and a full environment, installs dependencies, runs services and tests, and returns a pull request.

Developer costs $50 a month including $50 of usage. Model costs run through the customer's own provider accounts.

Roomote Roomote

Roomote is a source-available coding agent from the creators of Roo Code. It runs in an isolated, single-tenant deployment, with hosting and model choices under the team's control. It can investigate repositories, verify changes and return pull requests with live previews.

Teams can connect issue trackers, documentation and operational tools to supply context, then review the resulting changes through their usual pull request workflow.

[Self-hosting is free for up to 10 users](https://roomote.dev/). Self-hosted licenses are USD 249 per month for 11–50 users and USD 499 per month for 51–100 users, with larger deployments priced on request. Infrastructure and inference are paid directly to your providers.

Snowflake CoCo, formerly Cortex Code Snowflake CoCo, formerly Cortex Code

Snowflake CoCo has desktop and CLI agents for data engineering, analytics and AI work. Its CLI can orchestrate Snowflake skills and configured MCP tools, including tasks such as creating a dbt project or building a Streamlit application.

The CLI supports macOS, Linux and Windows. It requires a suitable Snowflake account or a dedicated CoCo CLI trial; standard Snowflake trial accounts do not provide access.

[The self-service subscription is USD 20 per month](https://www.snowflake.com/en/product/snowflake-coco/) after a 30-day trial with USD 40 in inference credits. Model consumption and Snowflake platform compute or storage are billed separately.

Tabnine Tabnine

Tabnine is part of Tricentis. Its current Tabnine Plugin for OpenCode combines the OpenCode agent with Tabnine sign-in, organisation governance, model access and built-in context tools. The provider is locked to Tabnine, and available models depend on the organisation's settings.

The older Tabnine CLI is in maintenance mode, with only critical updates through 31 December 2026 before full deprecation. Tabnine directs those users to the OpenCode plugin.

For an enterprise rollout, evaluate the plugin’s organisation controls and automatically registered code search and coaching tools. These are part of the Tabnine distribution that sits around OpenCode. The installation includes a managed copy of OpenCode, launched with the Tabnine command, so teams migrating from the previous CLI should use the documented plugin path.

[Pricing on request](https://www.tricentis.com/contact-us).

Tembo Tembo

Tembo runs coding agents in isolated cloud machines on existing repositories. Work comes from the dashboard, connected tools or automations that start fixes from errors or on a schedule; the agent edits files, runs commands and tests, and opens a pull request. Pro costs $60 a month with $90 of usage.

TRAE TRAE

TRAE comes from the team behind ByteDance's trae-agent. Its standalone SOLO app became TRAE Work on 9 June 2026. In Code Mode it takes a task in conversation, writes and runs code, and leaves the result for review.

The desktop app opens local project files; the web version runs tasks in the cloud.

Warp Warp

Warp offers an open source terminal and an Agent CLI that works in other terminals. The downloadable terminal remains one way to work with agents locally.

For teams evaluating broader automation, Warp Factories orchestrates agents across development workflows and is offered through early access. It supports a team's choice of models and agent harnesses, and can take work from tools such as GitHub, GitLab and issue trackers.

[Build is USD 20 per month](https://www.warp.dev/pricing), Max is USD 200 per month, and Business is USD 50 per user per month; annual billing lowers these to USD 18, USD 180 and USD 45 per month respectively. Free includes core terminal features and limited cloud-agent access. Enterprise pricing is on request, and additional agent usage and external model-provider charges are separate. · terminal · agent · rust

Zencoder Zenflow Code Zencoder Zenflow Code

Zencoder's Zenflow Code coordinates agents that build, test, review and refactor code. Teams can define workflow steps, run tasks in isolated environments and trigger background work from schedules, webhooks, pull request events or Jira updates.

Its verification workflow runs tests and cross-agent code review. Zencoder is operated by For Good AI Inc., and the product has a downloadable app for working with those agent workflows.

[Pro is USD 45 per user per month](https://zencoder.ai/pricing), Pro Plus is USD 95, and Pro Max is USD 195 per user per month; annual billing lowers these to USD 40, USD 85 and USD 175 per user per month respectively. Enterprise pricing is on request. The free Pro trial lasts up to seven days or 5,000 credits, and usage through your own provider keys is charged separately by the provider.