Primary question: Should you adopt a terminal AI agent that is actively being migrated to a successor repo?
RepoDaily adoption score
RepoDaily rates this as 91/100 (strong) for adoption: evidence, installation path, production risk, differentiation, license clarity, and AI/agent fit are scored from the article sources and adoption notes.
5 source(s) across 3 source category/categories, plus a RepoDaily-specific evidence module when available.
6 workflow step(s), 6 next-action step(s), and 2 command/install signal(s) were detected.
Trending momentum is +418 stars, with maintenance/release/issue signals counted when present.
Risk is marked medium, with 6 security note(s) and 4 explicit skip condition(s).
3 opportunity lens item(s), 4 alternative(s), and 3 type-specific section(s) support differentiation.
License source or license wording is present.
7 AI/agent-related signal(s) were detected in the article text and metadata.
Project overview
Kimi CLI is a Python-based AI agent that runs in the terminal, built by MoonshotAI. It reads and edits code, executes shell commands, searches and fetches web pages, and autonomously plans and adjusts actions during execution. The package is published on PyPI as kimi-cli and reached version 1.49.0 on 2026-07-16, four days before the trend date.
The agent distinguishes itself with a built-in shell command mode toggled by Ctrl-X, letting users run shell commands without leaving the Kimi CLI session. Beyond the terminal, it integrates with Visual Studio Code through a dedicated extension, with Zed and JetBrains IDEs via Agent Client Protocol (ACP), and with Zsh through a companion plugin called zsh-kimi-cli.
A prominent deprecation notice in the README states that Kimi CLI is evolving into Kimi Code CLI at github.com/MoonshotAI/kimi-code. Installing the successor automatically migrates configuration and sessions. The current project will be gradually wound down, though documentation and existing installations remain available. A /upgrade slash command was added in v1.47.0 to guide users through the migration.
On the security front, the project patches actively. Version 1.43.0 bumped pillow to 12.2.0 to address CVE-2026-25990, an out-of-bounds write triggered when loading PSD images. The SECURITY.md policy commits to supporting only the latest version, and vulnerabilities can be reported through the GitHub Security page or public issues. These factors make Kimi CLI worth evaluating for security-conscious terminal workflows, but the wind-down timeline adds a real adoption caveat.
Why it is trending now
- 418 period stars on 2026-07-20 at trending rank 11, driven by the v1.49.0 release four days earlier
- Version 1.49.0 shipped an adaptive completion-token budget that clamps max_completion_tokens to the model's remaining context window, reducing overflow errors on long turns
- Shell command mode (Ctrl-X) lets users run shell commands inside the agent session without switching windows
- ACP integration enables Zed and JetBrains IDE usage via the command kimi acp, broadening the tool's reach beyond the bare terminal
- Active CVE patching: pillow 12.2.0 for CVE-2026-25990 was shipped in v1.43.0, unblocking installs in environments that gate on older pinned versions
Problem it solves
- Developers juggling multiple terminal tabs to switch between a coding agent and a regular shell prompt
- IDE-bound AI agents that cannot run alongside terminal workflows without a protocol bridge like ACP
- Context-length overflow errors on long conversations when the CLI sends a fixed max_tokens value regardless of remaining context window
- MCP server configuration scattered across different tools, with no unified sub-command for adding and managing servers
- OAuth token persistence for MCP servers that relied on in-memory storage and required re-authentication on restart
How it works
- Install the package from PyPI under the name kimi-cli, as confirmed by the PyPI version badge in the README
- Launch Kimi CLI in the terminal and send /login to complete authentication before using any agent features
- Toggle shell command mode with Ctrl-X to run shell commands directly within the session. Built-in shell commands like cd are not yet supported
- For IDE integration, configure ACP-compatible clients (Zed or JetBrains) to launch Kimi CLI as an ACP agent server using the command kimi acp, with the JSON config block shown in the README placed in ~/.config/zed/settings.json or ~/.jetbrains/acp.json
- Manage MCP servers via the kimi mcp sub-command group, for example: kimi mcp add --transport http context7 https://mcp.context7.com
- Run /upgrade when ready to migrate to Kimi Code CLI, which automatically moves configuration and sessions to the successor project
Product demo and interface preview
Command and Slash-Command Surface
- kimi acp — starts Kimi CLI as an ACP agent server for IDE clients
- kimi mcp add --transport http context7 https://mcp.context7.com — adds a streamable HTTP MCP server
- kimi mcp auth <name> — re-authenticates an OAuth MCP server after upgrading to FastMCP 3 storage in ~/.kimi/mcp-oauth/
- /login — completes authentication before any agent or ACP usage
- /upgrade — installs Kimi Code CLI and migrates config and sessions (added in v1.47.0)
- /clear or /new — starts a new session; /clear became an alias for /new in v1.45.0
- /help (alias /h or ?) and /usage (alias status) — slash-command aliases resolved since v1.44.0
- Ctrl-X — toggles shell command mode inside the agent session
- KIMI_MODEL_MAX_COMPLETION_TOKENS env var — sets an explicit hard cap on completion tokens (KIMI_MODEL_MAX_TOKENS remains a compatibility alias; 0 or negative disables clamping)
Integration Surface
- VS Code — via the Kimi Code VS Code Extension on the marketplace (extension ID: moonshot-ai.kimi-code)
- Zed — via ACP, configured in ~/.config/zed/settings.json under agent_servers with command kimi and args ["acp"]
- JetBrains IDEs — via ACP, configured in ~/.jetbrains/acp.json with the same agent_servers block
- Zsh — via the zsh-kimi-cli plugin, cloned into ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/kimi-cli and added to the plugins list in ~/.zshrc; after restart, Ctrl-X switches to agent mode
- MCP servers — managed through kimi mcp sub-command group, with OAuth tokens persisted in ~/.kimi/mcp-oauth/ via FastMCP 3 storage API
Maintenance Risk and Wind-Down
- README deprecation notice: Kimi CLI is evolving into Kimi Code CLI at github.com/MoonshotAI/kimi-code; config and sessions migrate automatically via /upgrade
- SECURITY.md: only the latest version receives security support — no backports to older releases
- CONTRIBUTING.md: pull requests larger than 100 lines require prior discussion via an issue or risk being closed without review
- Pre-commit hooks use prek (github.com/j178/prek); developers run make prepare to sync dependencies and install hooks
- v1.42.0 switched the Windows shell backend from PowerShell to Git Bash, running commands through bash.exe
Who should pay attention?
Good fit if
- Developers who live in the terminal and want an AI agent that also runs shell commands in the same session
- Teams using Zed or JetBrains IDEs who want to bring a custom agent via ACP without a dedicated plugin
- Zsh users who want agent-mode capabilities inside their existing shell via the zsh-kimi-cli plugin
- Anyone evaluating MCP-compatible toolchains who needs a CLI agent with first-class kimi mcp management
- Security-conscious developers checking how a terminal agent handles CVE patching and version support scope
Skip for now if
- Teams that require long-term support for older versions — SECURITY.md supports only the latest release
- Organizations that cannot accept a wind-down timeline — the project is migrating to Kimi Code CLI
- Windows users who depend on PowerShell as the shell backend — v1.42.0 switched to Git Bash
- Anyone who needs built-in shell commands like cd to work inside shell mode — the README notes they are not yet supported
Risks and cautions
The tool is functional and actively patched, but the README declares a wind-down in favor of Kimi Code CLI, and SECURITY.md limits support to the latest version only.
- README deprecation notice explicitly states the project will be gradually wound down
- Only the latest version receives security support per SECURITY.md — no LTS or backport commitment
- The /upgrade command added in v1.47.0 is designed to migrate users away from this repo
- Contributions larger than 100 lines require prior maintainer discussion per CONTRIBUTING.md
- SECURITY.md: only the latest version is supported; no patches for older releases
- v1.43.0 bumped pillow to 12.2.0 to fix CVE-2026-25990 (out-of-bounds write when loading PSD images)
- OAuth MCP tokens stored in ~/.kimi/mcp-oauth/ using FastMCP 3 persistent storage API since v1.43.0
- MCP client stack upgraded to FastMCP 3.2.4 to eliminate Authlib deprecation warnings during OAuth startup
- Shell command mode (Ctrl-X) executes arbitrary shell commands within the agent context — review before deploying in restricted environments
- Vulnerabilities reportable via the GitHub Security page or public issues, as documented in SECURITY.md
Alternatives to compare
| Approach | When to use | Trade-off |
|---|---|---|
Aider | You want an open-source CLI coding agent focused on Git-integrated pair programming without an ACP layer | Free, open-source; bring your own API keys |
Open Interpreter | You need a terminal AI that executes code locally across multiple languages with broader OS-level control | Free, open-source; bring your own model access |
Claude Code | You want a commercial terminal agent backed by Anthropic with enterprise support | Commercial subscription required |
Kimi Code CLI (successor) | You are starting fresh and want the actively developed next-generation terminal agent from the same team | Free to install; same Kimi account model |
What this trend reveals
ACP-compatible IDE workflows
Kimi CLI ships ACP support out of the box, so any editor implementing Agent Client Protocol can use it as a custom agent server. Developers using Zed or JetBrains can configure kimi acp without waiting for a dedicated plugin.
Verify by adding the agent_servers JSON block to ~/.config/zed/settings.json or ~/.jetbrains/acp.json and creating a Kimi CLI thread in the IDE agent panel.
MCP toolchain consolidation
The kimi mcp sub-command group lets you add, manage, and authenticate MCP servers from one CLI. OAuth tokens persist in ~/.kimi/mcp-oauth/, eliminating re-login cycles after restarts.
Run kimi mcp add --transport http context7 https://mcp.context7.com and confirm the server appears in subsequent sessions without re-authentication.
Controlled migration to the successor
The /upgrade command automates config and session migration to Kimi Code CLI, reducing switching cost for existing users and providing a clean exit path during the wind-down.
Run /upgrade in a test session and verify that configuration and session history carry over to the Kimi Code CLI installation.
RepoDaily verdict
Kimi CLI delivers a capable terminal AI agent with shell mode, ACP IDE integration, and MCP support — but the README's wind-down notice and SECURITY.md's latest-version-only policy mean new adopters should plan for migration to Kimi Code CLI rather than long-term use of this repo.