Docs · BoltLynx

Quick start

Set up the machine or isolated environment where you want Agents to act. One local client can serve multiple Agents.

Install the local client.
Start the local client and confirm the pairing page in your browser.
Open the web app and create or select an Agent.
Choose the project folder the Agent is allowed to use.

Windows local client

Download the Windows launcher:

Create a permanent folder you will keep, such as C:\BoltLynx, then extract all files from the ZIP into it. This is a portable build; no installer is required.

Do not run boltlynx-desktop.exe from inside the ZIP window. Windows may start it from a temporary folder, making the launcher and its local files difficult to find later.

From the extracted folder, double-click boltlynx-desktop.exe. The Windows launcher manages the local client; it is not where you talk to Agents. Use it to start or stop the client and to open or copy the pairing page.

Windows may show a Microsoft Defender SmartScreen warning because this early desktop build is not signed yet. If you downloaded it from the official button above, click More info, then Run anyway.

If antivirus software flags the Windows launcher

The local client can perform system, command, and file actions that you authorize. Unsigned builds may therefore trigger heuristic antivirus detections such as Trojan:Win32/Sprisky.U!cl. Do not allow a detection automatically.

First confirm that the ZIP came from the official download button above and that the detected file is C:\BoltLynx\boltlynx-desktop.exe. If the source, file name, or detected file does not match, delete it and download a fresh copy from the official button.
If those details match, open Windows Security → Virus & threat protection → Protection history, open the detection, and choose Allow on device. If the file was quarantined, choose Restore. The wording may vary slightly by Windows version.
If Defender repeatedly quarantines the confirmed file, open Virus & threat protection settings → Manage settings → Exclusions → Add or remove exclusions, choose Folder, and add only C:\BoltLynx.

Do not turn off real-time protection, and do not exclude Downloads, your whole user folder, or an entire drive.

Microsoft instructions: Protection History and folder exclusions.

After pairing, leave the local client running in the background. Open the web app to talk to Agents and choose the project folder they are allowed to use.

When an update is available, the Windows launcher can download the new local client. Because the launcher is portable, its local files stay under the folder where you run the executable.

Linux command line

Install the local client:

curl -fsSL https://boltlynx.com/install.sh | bash

The installer places the boltlynx binary under ~/.boltlynx/bin/ and adds it to your PATH. Restart your terminal, then start the local client. On first run, the command shows a link to the pairing page; open it in your browser and confirm. The local client then keeps running in the background. No separate pairing command is needed.

boltlynx start

Check the client status at any time:

boltlynx status

macOS command line

Install the local client:

curl -fsSL https://boltlynx.com/install.sh | bash

boltlynx start

Check the client status at any time:

boltlynx status

Once the local client is running and paired, open the web app, create or select an Agent, and choose the project folder the Agent is allowed to use.

Open the web app

Upgrade the local client

After an update:

Windows: update the local client from the launcher, then restart the client kernel.
Linux / macOS: restart the local client from the terminal.

Restart the local client

boltlynx restart

Core concepts

Web app

Talk to Agents, create and manage work, and decide what should happen next. The web app runs in your browser.

Local client

The client runs on your machine. It gives approved agents access to local files, commands, data analysis, code search, and private skills.

A client is not one Agent. It is an isolated runtime environment — a machine, a container, or a separate OS user on the same machine. Install the client once for that environment; it can host multiple Agents.

Agent

An agent is a conversation plus a working environment. It has conversation history, a working directory, memory, enabled skills, and its own instructions.

Skills

Skills give an Agent domain abilities. The current guide covers Code for software work and Ops for operations and database safety rules.

Worker agents

A main agent can delegate work to specialized worker agents. Each worker can have its own model, working directory, skills, and role.

Agent setup

Each agent can be configured for a specific project or workload.

Name, description, working directory	Identify the agent and define the root folder it can operate in.
Skills	Enable domain abilities such as Code or Ops.
Model	Choose the model and reasoning level for the work.
File permissions	Set allowed, writable, or blocked paths.
Command permissions	Allow or block command patterns, such as allowing `git *` while blocking destructive commands.
Instructions and memory	Give the agent long-term working rules and editable notes it can carry across conversations.
Worker agents	Configure specialized collaborators for larger tasks.
Triggers and notifications	Run scheduled checks, webhook-triggered workflows, or send completion and alert notifications.

Daily use

Open the web app. Select the Agent for your project, then continue the conversation or give it the next piece of work.
Describe the work. Give the agent a real task, not just a prompt.
Let it operate. The agent can read and write files, run commands, search code, and analyze data inside the approved environment.
Approve sensitive actions. Operations that require confirmation ask for human approval first.
Get notified. BoltLynx can send notifications when work completes, when anomalies appear, or when scheduled tasks report back.

Safety model

BoltLynx is designed around explicit boundaries. The agent works inside the directories and command permissions you configure.

File access is permissioned by working directory and additional allow/block rules.
Sensitive commands can require human approval.
Private skills run through your local client instead of being uploaded as server-side tools.
Notifications can surface completion, failures, and scheduled reports.

Data boundaries

BoltLynx is designed to minimize both local access and data transfer. It does not mirror your workspace or scan and upload unrelated files.

Work content leaves the client only when needed for model reasoning, required action results, an online service you choose, or operational diagnostics needed to run and secure the Service.

Install the local client. Start real work.