Cline is an open-source AI coding agent, published by Cline Bot Inc. under the Apache License 2.0, that reads and edits files, runs terminal commands and uses tools in your editor, terminal, desktop app or a cloud sandbox, asking for approval before each action unless you allow it to proceed. The client is free; what people pay for is model inference, either through their own provider keys, Cline's usage-billed credits or the ClinePass subscription. That split confuses many buyers, so this page sets out what Cline is, how the costs work, which models and providers it reaches, its features and approval settings, its enterprise controls, how data flows and how the product has changed. It is written for developers, engineering managers and the people who approve tools for a team. Every fact comes from Cline's website, documentation, GitHub repository and changelog. Last checked: 2026-11-15.
This is an independent guide from NHI Mgmt Group, based on Cline's public documentation as of November 2026. It is not sponsored or reviewed by Cline. Product features change, so confirm settings against Cline's current documentation.
Key takeaways
- Open source, Apache 2.0. The cline/cline repository carries the Apache License 2.0; the client has no subscription or seat fee in its open-source form.
- You pay for inference, three ways. Bring your own key, buy Cline credits billed by usage, or subscribe to ClinePass at $9.99 a month for higher limits on open-weight coding models.
- Many surfaces, one agent. A VS Code extension, a JetBrains plugin, a CLI, a desktop app, an SDK, Kanban and browser-based Cloud Agents share the same agent and Plan and Act modes.
- Approval is configurable per action type. Auto-approve toggles cover reads, edits, commands, browser and MCP use; YOLO mode approves everything and the docs label it dangerous.
- Data path depends on the provider. With your own keys, prompts go straight to your provider; with Cline or ClinePass inference they pass through Cline, which states that code is never used to train models.
What is Cline?
Cline's documentation describes it as an AI coding agent that runs in your editor and terminal, can read and write files, run commands and use a browser, and "asks for your approval before each action". The home page summarises the positioning as "Open source. Plan-and-Act. Every model. No vendor lock-in."
The project began as a VS Code extension. Its changelog records version 2.0.0 with the line "New name! Meet Cline, an AI assistant that can use your CLI and Editor", and later milestones add MCP support (2.2.0), a root-level .clinerules file and granular auto-approve (3.0.0), and an SDK-backed runtime with ClinePass and a Customize marketplace (4.0.0). Cline's blog states that the move of the VS Code extension to the new SDK covered 11 million developers, and the home page says it is trusted by over 11 million developers.
Cline Bot Inc. holds the copyright on the website. Cline's own pages do not state a founding date or funding details, so none are given here.
Key facts at a glance
As of 15 November 2026, from Cline's own pages and repository.
| Field | Value | Source page |
|---|---|---|
| Publisher | Cline Bot Inc. | cline.bot |
| Licence | Apache License 2.0 | GitHub LICENSE file |
| Product type | Open-source coding agent: IDE extensions, CLI, desktop app, SDK, Kanban, Cloud Agents | Docs home |
| Latest extension version | 4.1.26 (GitHub release listed 26 October) | Changelog, releases |
| Other release tracks | CLI v3.0.74, Desktop v0.0.49, SDK v0.2.0 | GitHub releases |
| Client price | Free; pay for inference | Pricing page, FAQ |
| ClinePass | $9.99 per month | ClinePass docs |
| Enterprise | Custom pricing via sales | Pricing page |
| Model access | Cline provider, ClinePass, ClineFree models, or your own key for cloud or local providers | Docs, Cloud Agents docs |
| Training on customer code | Cline states it never uses code or prompts to train models | FAQ, Enterprise page |
| Certifications | Not stated by Cline on the pages checked | Not stated |
Plans and pricing
Cline's pricing page lists two plans, and the documentation adds the inference options that sit beside them. Prices are as shown on Cline's pages on 15 November 2026; the pricing page does not state a currency.
| Option | Price | What you get |
|---|---|---|
| Free (open source) | No subscription or seat fee | VS Code extension, CLI, client-side architecture, inference at cost or bring your own key, MCP Marketplace, multi-root workspaces, community support |
| Cline provider (usage billing) | Pay as you go with Cline credits | One sign-in and built-in billing for frontier and open models, plus free model options |
| ClineFree models | $0 | Free models shown as "(free)" in the picker, with a reset time when the free limit is reached; the list changes over time |
| ClinePass | $9.99 per month | Flat subscription for open-weight coding models with 2 to 5 times the usage of standard API rates; limits on a 5-hour rolling window, a weekly limit and a monthly limit |
| Bring your own key | Your provider's prices | Your credentials for cloud providers or local runtimes; requests go straight to the provider |
| Enterprise | Custom | Everything in Free plus JetBrains extension, SSO, SLA, dedicated support, centralised billing, config management, role-based access control, inference provider limits, team dashboard, authentication logs |
Plan table versus FAQ: the pricing page lists the JetBrains extension under Enterprise, while the docs home lists a JetBrains plugin among the general applications. The pricing page also marks Advanced Config Management and Fine-Grained Permissioning as "Coming Soon". Check the current pricing page before relying on either.
What this means for budgeting: with your own keys, Cline adds no charge and cost equals provider usage, so provider-side spending limits do the budgeting. ClinePass gives a fixed monthly cost but only covers its listed open-weight models. The ClinePass docs show per-million-token prices only as a reference for how usage counts against the quota.
Models available
Cline is provider-agnostic, so the model list is long and moves with every release. The home page lists Claude, GPT and Gemini models, local models through Ollama or LM Studio and any OpenAI-compatible endpoint; the enterprise docs name Amazon Bedrock, Google Vertex AI, Azure OpenAI, Anthropic, OpenAI and Cerebras among supported providers, and the docs cover more than 30 further providers.
| Route | Models as documented on 15 November 2026 | Billing |
|---|---|---|
| Cline provider | Frontier and open models; with no model chosen, release 4.1.26 defaults to Claude Sonnet 5.5 | Cline credits, usage based |
| ClinePass | 15 open-weight models including DeepSeek V4.1 Flash and V4 Pro, GLM 5.3 and 5.3 Flash, Kimi K3 and K3.1 Pro, MiniMax M3 and M3.1 Flash, MiMo V2.6 Pro and Flash, Muse Spark 1.3 Contributor, Qwen3.7 Max and Plus, Qwen3.8 Max; release 4.1.26 sets Kimi K3.1 Pro as the default | $9.99 per month subscription |
| ClineFree | A rotating list; release 4.1.26 adds Reflection Beam, Nemotron 3.5 Lightning and Laguna S 2.1 | Free, with limits |
| Bring your own key | Any model your provider or local runtime serves, including OpenAI-compatible endpoints | Provider bills you |
- Per-mode models: turning on "Use different models for Plan and Act" lets you pick a stronger reasoning model for planning and a faster one for implementation; switching modes switches models.
- Defaults move: the 4.1.26 changelog warns that default models changed for most providers, so "if you use a provider without pinning a model, expect a different default". Pin the model in team guidance.
- Context windows: Cline does not publish its own figure; the limit is the selected model's, as stated by its provider.
Features and modes
| Feature | What it does (per Cline docs and changelog) | Where it runs |
|---|---|---|
| Plan and Act | Plan mode reads, searches and discusses but cannot modify files or run commands; Act mode carries out the plan with the same context | All surfaces |
| Multi-file edits and checkpoints | Coordinated changes with diffs, linter-aware fixes, checkpoints and undo at each step | Editor, desktop |
| Terminal commands | Runs commands and reacts to output, including dev servers, tests and deployments | Editor, CLI, desktop |
| Auto-approve and YOLO | Per-category approval toggles; YOLO approves all actions including Plan to Act transitions | Editor, desktop |
| Rules and skills | .clinerules files encode standards and conventions; skills and slash commands package reusable tasks | All surfaces |
| MCP and plugins | MCP servers for databases, APIs and infrastructure; MCP Marketplace; Plugins and the Customize marketplace (4.0.0); hooks and custom tools through the SDK | Editor, desktop, SDK |
| Integrations | Chat with the agent from Slack, Discord, Telegram or Linear; headless runs in GitHub Actions, GitLab and other pipelines | CLI, cloud |
| Multi-agent and scheduling | Coordinator agents delegate to specialists; tasks can run on a cron schedule; Kanban board for parallel agents | Desktop, CLI |
| Cloud Agents | Browser-started tasks on a GitHub repository in an isolated sandbox, delivered as a branch and pull request on a cline/ branch | Cline cloud |
| SDK | @cline/sdk embeds the agent runtime in your own apps, with tool permission handling | Your code |
Auto-approve toggles: Read project files, Read all files, Edit project files, Edit all files, Execute safe commands, Execute all commands, Use the browser, Use MCP servers and Enable notifications. The docs recommend turning on Read project files and leaving the rest off. There is no fixed command allowlist: the model sets a requires_approval flag for each command, and the docs state that their examples of safe commands are not guarantees.
Supported IDEs and platforms
- VS Code extension: the original surface, also usable in VS Code-based editors; the home page lists Cursor and Windsurf.
- JetBrains: a plugin for IntelliJ IDEA, PyCharm, WebStorm and other JetBrains IDEs.
- ACP mode: editors that speak the Agent Client Protocol, including Zed, Neovim and Antigravity per the docs.
- CLI: installed with
npm i -g cline, interactive or headless for CI/CD and scripting. - Cline Desktop: an open-source app that needs no editor, for parallel sessions and scheduled tasks, announced on Cline's blog on 14 September 2026; the home page offers macOS and Windows downloads and lists Linux.
- SDK and Kanban:
npm install @cline/sdkandnpx kanbanfor a web task board of parallel agents. - Cloud Agents: from app.cline.bot in a browser, with up to 10 concurrent sessions and 50 new sessions per day.
Enterprise and admin controls
| Control | Documented setting or feature |
|---|---|
| Single sign-on | SSO with corporate credentials in place of personal API keys |
| Roles | Member, Admin and Owner, scoped to the organisation |
| Model and tool governance | Per-team model and tool access; inference provider limits; MCP controls |
| Remote configuration | Global settings for all developers from one dashboard, including organisation telemetryEnabled and YOLO mode controls |
| Audit | Authentication logs; selective audit logging of administrative operations |
| Observability | OpenTelemetry export to Datadog, Grafana and Splunk; usage and cost by team and model; Prompt Storage backup of conversation history to AWS S3 or Cloudflare R2 |
| Deployment | Enterprise VPC or inference host including Amazon Bedrock and Google Vertex; the enterprise page lists VPC, on-premises and air-gapped deployment |
The pricing FAQ also names OIDC, SCIM provisioning and VPC deployments for Enterprise, which the plan table does not list; confirm scope with Cline before contracting.
Provider API keys stored in each developer's Cline settings are non-human credentials with real spend and data access. Moving to SSO and centrally configured providers removes personal keys from laptops. See AI Coding Tools and Secrets Sprawl and How to audit AI coding assistant tokens and config files.
Data handling and privacy
- Your own keys: the client runs locally and requests go directly from your environment to the chosen provider. The enterprise docs add that Cline does not index or embed repository contents and reads files only when a task needs them.
- Cline or ClinePass inference: code and prompts pass through Cline's infrastructure to reach the model; Cline states it sees requests only to route them and that code is never used to train models.
- Telemetry: usage and error events, token counts, model and provider names, version, OS and editor, with a device ID; code, file contents and prompts are excluded except for requests to the Cline and ClinePass providers, where telemetry can include prompts and responses.
- Turning telemetry off: in VS Code and JetBrains clear "Allow error and usage reporting" under General; in Cline Desktop turn off Telemetry under Settings > General; in the CLI set
"telemetryOptOut": truein~/.cline/data/settings/global-settings.json. The setting is shared across Cline products on one computer. - Enterprise override: an organisation can lock telemetry off for VS Code and JetBrains members through remote configuration; the docs state that Cline Desktop and the CLI ignore this organisation setting.
- Retention: Cline's pages checked do not state retention periods.
Release history
Cline's changelog lists versions without dates, so early milestones are given by version only. Dated items come from Cline's blog and GitHub releases page.
| Version or date | Release or event | Headline changes |
|---|---|---|
| 2.0.0 | Renamed to Cline | New name, streamed responses, cancel button, model choice through OpenRouter |
| 2.2.0 | MCP | Model Context Protocol support and MCP server management |
| 3.0.0 | Rules and approvals | Search and replace diffs for large files, granular auto-approve, notifications, root .clinerules |
| 4.0.0 | SDK runtime | SDK-backed VS Code runtime, ClinePass in the extension, Customize marketplace for Skills, MCP servers and Plugins, queued prompts |
| 4.0.12 | Free models | Free Cline models marked "(free)" with reset times |
| 4.1.0 | A/B package | One VSIX carrying the legacy and SDK-based extensions with a staged rollout loader |
| 2 September 2026 | Harness migration post | Blog describes moving 11 million users to the SDK harness |
| 14 September 2026 | Cline Desktop | Open-source desktop app for parallel agents, recurring tasks and 300+ models |
| 4.1.26 (26 October) | Cloud tasks in the extension | Local and Cloud toggle, sandbox progress card, retries cut from 5 to 3, model catalogue refresh |
Limits and known constraints
- Command safety is model-judged: "Execute safe commands" relies on the model's
requires_approvalflag, not a fixed allowlist. - YOLO mode: has no confirmation dialog and the docs list outcomes including deleted files, changed system settings and pushed commits.
- ClinePass scope and limits: covers only its listed open-weight models, metered on 5-hour, weekly and monthly limits; models are added and removed between releases.
- Cloud Agents: up to 10 concurrent and 50 new sessions per day; idle sessions pause after 15 minutes; rules, skills, MCP servers and connectors do not yet load in cloud sessions.
- Telemetry lock coverage: the organisation telemetry setting applies to VS Code and JetBrains but not to Cline Desktop or the CLI.
- Roadmap items: Advanced Config Management and Fine-Grained Permissioning are marked "Coming Soon" on the pricing page.
Security guidance
This page describes what Cline is and costs. For hardening, apply the discipline in the AI IDE and Coding Agent Security Guide. The Cline-specific points: keep auto-approve to Read project files for work repositories, and block YOLO mode centrally through remote configuration where you run Enterprise. Review MCP servers and plugins before installation, since they extend what the agent can reach; see the MCP Security Guide. Treat .clinerules, skills and other repository-supplied instruction files as untrusted input in cloned code; see Coding Agent Configuration File Attacks. Because Cline is open source under Apache 2.0, check redistribution or internal forks against your policy using the Open Source Licence Compliance Guide.
Practitioner checklist
- Decide which inference route is allowed (own keys, Cline credits, ClinePass or enterprise VPC) based on where prompts and code may travel.
- Pin the model per provider in team guidance, since release defaults change.
- Set auto-approve to Read project files only, and disable YOLO mode through remote configuration on Enterprise.
- Replace personal provider keys with SSO and centrally configured providers, and inventory any keys that remain, with an owner and rotation date.
- Decide on telemetry, remembering that Cline and ClinePass provider telemetry can include prompts, and that the organisation lock does not cover Desktop or the CLI.
- Export OpenTelemetry to your SIEM and review authentication and admin audit logs.
- Limit Cloud Agents GitHub access to the repositories that need it and require review on
cline/branch pull requests. - Recheck the pricing page and changelog before renewal or when standardising guidance.
Standards and references
- Cline: Documentation
- Cline: Pricing
- Cline: ClinePass
- Cline: Auto-approve
- Cline: Enterprise security and governance
- Cline: Telemetry
- Cline: LICENSE (Apache License 2.0)
- Cline: Changelog
Related NHI Mgmt Group resources: AI IDE and Coding Agent Security Guide · AI Coding Tools and Secrets Sprawl · MCP Security Guide · AI Coding Assistant Buyer's Guide