Privateer launches open-source terminal AI agent

Privateer has launched Privateer CLI, an open-source AI coding agent for the terminal that the company says can read, write, and run code across projects. The announcement appeared in a post on X on October 5, 2026, and the company says the product is also available on GitHub and npm as privateer-agent.

The release targets builders who want an agentic coding workflow without tying their projects to a single model provider. Privateer describes the CLI as a provider-agnostic private AI coding agent, and says tool-calling and streaming work the same across providers, with no model lock-in or separate code paths. The company also says users can switch models with /model.

What does Privateer CLI do?

Privateer CLI is designed to sit in the terminal and handle coding tasks with user approval gates in front of actions. The company says the agent can read files, write code, and run code, but every action is classified and must be approved before it executes. Privateer says the permission system also intercepts tools added by packages, including MCP tools, and blocks destructive shell commands even in unattended runs.

The product page says the CLI includes attested TEE inference, an on-device PII gate, MCP servers, sub-agents, scheduled routines, chat integrations, and remote control from Android or the web. Privateer also says the CLI is built on the open Pi agent toolkit.

For founders and solo builders, that means the product is aiming at a broad "agent in the terminal" workflow rather than a narrow code-completion tool. Privateer says the CLI can launch in any project with the privateer command or with npx privateer-agent, and that users with Node 22+ can also run it through npm. The company says the self-contained installers for macOS, Linux, and Windows bundle their own runtime, so users do not need Node or npm for the installer path.

How does Privateer handle models and deployment?

Privateer says users can bring their own model, use a Privateer account, or run locally with no key at all. In the announcement post, the company said the terminal coding agent works with a user's own model, an account, or locally with no key, and that it is open source under the MIT license.

On the product page, Privateer says one provider key is enough to start, but users can also sign in and bill through a Privateer account instead. The company lists support for major providers, plus local Ollama and any OpenAI-compatible endpoint. It also names a long list of supported bring-your-own-key providers, including OpenRouter, Anthropic, OpenAI, Google, xAI, Groq, Mistral, Z.ai, DeepSeek, Together, MiniMax, Qwen, Ollama, NEAR AI, Tinfoil, Venice, Fireworks, and custom endpoints.

Privateer says local Ollama can run inference on the user's own machine with no API key. It also says the CLI can use /signin to proxy prompts through its service without storing them, while keeping billing metadata such as model, token counts, and cost. According to the company, local PII detection is deterministic and never leaves the user's machine.

The company says NEAR AI and Tinfoil run models inside attested trusted execution environments, and that /verify can fetch a live cryptographic attestation. For users who want to avoid model-specific integrations, Privateer says its tool-calling and streaming layer behaves consistently across providers.

What cross-platform and collaboration features are included?

Privateer is also packaging the CLI with remote access and chat controls. The company says remote access is off by default and can be enabled with /remote-access to link a terminal to the Android app or web app, while execution stays on the user's machine. It says the remote relay is live-only, nothing is archived, and no keys cross the relay.

The product page says users can bridge a terminal to Telegram, Slack, Discord, or WhatsApp and message the agent like a teammate. Privateer says chat-connected admins can approve actions in chat while everyone else is read-only, and that per-channel posture can be set to read-only, approve-each-action, or unattended. Bot tokens for those integrations are sealed to the user's machine, and configuration is signed by the user's account.

Privateer also says the CLI supports routines, which are saved tasks run by a background daemon on a cron schedule or one-off basis. Routine outputs can be delivered to a file, the next session, a phone, email, or a webhook such as Slack or Discord, with secrets redacted before anything leaves the machine.

That set of features suggests Privateer is aiming beyond local coding assistance and toward a broader operational assistant for developers and small teams. The company’s framing centers on privacy controls, provider flexibility, and command-line workflows, all wrapped in an MIT-licensed open-source package.

What happens next for users?

For users, the immediate next step is simply choosing how to run it: with a provider key, through a Privateer account, locally with Ollama, or with no key at all, depending on the setup they want. Privateer says the project is available through GitHub, npm, the privateer command, and npx privateer-agent.

The company has not announced pricing details in the materials reviewed here, but it does say the product can be used without a key and that account mode keeps only billing metadata. For now, the launch positions Privateer CLI as a cross-platform, open-source terminal agent built for developers who want model choice, local execution options, and approval-based controls in one tool.