# Agents · Neural Inverse Build > A desktop app for running agent sessions in parallel and reviewing their changes, under the same policy as the rest of your team. Source: https://neuralinverse.com/build/agents/ --- ![](/photos/facility.jpg) [Neural Inverse Build / Agents](https://neuralinverse.com/build/) # Delegate work to agents. A person approves what matters. A desktop app for running agent sessions in parallel and reviewing their changes, under the same policy as the rest of your team. Part of Neural Inverse Build Platforms macOS, Windows and Linux Engines Claude Code, Codex and our own Sessions Several in parallel Record Every session, prompt and tool call Overview The Agents app is where engineers hand work to AI agents. Each session runs in a shared agent host under your organisation's policy, every action is decided before it runs, and every change waits as a diff for a person to review and keep. [What it does ↓](https://neuralinverse.com/#capabilities)[Workflow ↓](https://neuralinverse.com/#workflow)[Governance ↓](https://neuralinverse.com/#governance)[Technical facts ↓](https://neuralinverse.com/#technical)[Get started ↓](https://neuralinverse.com/#get-started) What it does ## What the Agents app does. Several agents at work at once, each one governed and each one on record. - **Parallel sessions:** Start a session in a workspace, choose a model and describe the task. Start more while the first one works; each is listed with its own chat and changes. - **Review changes:** The Changes tab shows every file a session changed, as diffs, with a detail panel for the branch's changes and checks. Keeping a change is a governed action of its own. - **Choice of engine:** Sessions run in a shared agent host that can run several engines: Claude Code, Codex and Neural Inverse's own engine. - **Policy translated per engine:** For Claude Code, organisation rules become its permission rules, bypass mode is turned off and denied MCP servers are removed. For Codex, rules set its approval policy, sandbox mode and network access. - **Decided before each call:** Whatever an engine's own settings cannot express is decided live, before each tool call. - **Stated enforcement:** Each engine version has a table of what it can enforce. A rule it cannot enforce is reported, never dropped, and a mandatory one blocks that engine for the session. - **Recorded without modifying engines:** Every engine's sessions, prompts and tool calls are recorded in the device's audit ledger through the engine's own hooks. - **Automations:** Schedule agent sessions on an agent host, or start them by hand. Key workflow ## Start, run in parallel, approve, review, keep. 1. Sign in To your organisation's Govern deployment, with your own account. 2. Start a session Choose a workspace and a model, and describe the task. 3. Run in parallel Start more sessions while the first one works. 4. Approve Actions your policy marks wait for a person before they run. 5. Review Open Changes and read every file the agent changed, as diffs. 6. Keep Accepting the changes is decided by the same governance service. Who uses it ## Built for every role on the team. - **Engineers:** Delegate tasks, run several sessions at once and review each diff before keeping it. - **Security teams:** Set which engines may run and what they may call, read, write and reach, through one signed policy. - **Platform teams:** Roll the app out with managed policy, using the same policies as the Editor. - **Compliance teams:** Read a record of every engine's sessions, prompts and tool calls, and the approvals behind them. How it is governed ## Every agent action is decided before it runs. One decision point for every engine the agent host runs, and it wins over each engine's own settings. - **Allow, deny or needs approval:** The decision point decides each action: the tool, each shell command after full parsing, each file read or written, each host contacted and each MCP server. - **Approvals:** When an action needs approval you are asked before it runs. Where an approval rule such as separation of duties applies, other enrolled approvers must sign. - **Built-in protection:** Even with no organisation policy, an agent cannot change policy and approval records, git hooks, CI configuration, shell profiles or audit data without a person. - **Signed policy on every engine:** The organisation's signed policy is verified on the device and applied to every engine. When it lapses past its offline grace, agent sessions, tool calls and model calls lock. - **Audit on the device:** Agent events stay in the device's ledger unless your organisation turns on uploading to its own Govern deployment. - **Stated limits:** A modified build, or an engine run outside the agent host, is not covered by the device's controls. What must hold is enforced by Govern and your CI. Technical facts ## For your IT and security review. - **Platforms:** macOS, Windows and Linux. Named Neural Inverse Agents on your system; installed with the Editor, with its own process, data folder and sign-in session. - **Sign-in:** Your organisation account on your organisation's Govern deployment: in the browser with OAuth 2.0 and PKCE, or with a device code (RFC 8628) on a machine with no browser. The session is kept in the operating system's credential store. - **Its own client:** The Agents app signs in with its own OAuth client, so its sessions, its entry in your device list and its audit records are its own. - **Managed policy:** The same managed policies as the Editor, including the address of your Govern deployment. - **Models:** Your organisation's models through Govern with your own sign-in, providers your organisation allows, or local models. Shared with the Editor. - **Without internet access:** Local models need no key, Govern can be self-hosted on premises, and policy is decided on the device with no Neural Inverse server called. No telemetry is sent to Neural Inverse. - **Licence and status:** Proprietary, under your agreement with Neural Inverse; source review under NDA. Pre-release: interfaces and data formats may change between versions. Get started ## Install, sign in and read the documentation. - [Install the Editor and Agents app](https://neuralinverse.com/docs/get-started/install/) - [Sign in to your organisation](https://neuralinverse.com/docs/get-started/sign-in/) - [Agents documentation](https://neuralinverse.com/docs/agents-app/) - [Models](https://neuralinverse.com/docs/build/models/) - [Policy, approvals and audit](https://neuralinverse.com/docs/build/governance/) Related [Editor → Hands-on development with AI you review line by line.](https://neuralinverse.com/build/editor/)[CLI → An AI coding agent for the terminal and for CI.](https://neuralinverse.com/build/cli/)[Neural Inverse Govern → Sign-in, approved models, policy and audit for every AI tool. Powered by Keystone.](https://neuralinverse.com/govern/) ![](/photos/team.jpg) ## Talk to our team. Neural Inverse is available by request. We work with your engineering, security and procurement teams from the first conversation. [Request access](https://neuralinverse.com/contact/?topic=access)[Contact us](https://neuralinverse.com/contact/)