# Introducing Excession

Published 2026-11-01 by Excession Labs · Source page: https://excessionlabs.ai/blog/introducing-excession/

> Excession is an AI agent for offensive security that runs in your terminal. A planner splits the job, workers run it in parallel, and every finding comes back with proof.

## What it is

Excession is a terminal workspace for authorized security work: reconnaissance, web and API testing, code audit, exploit development, reverse engineering and malware analysis. It is its own harness. Excession owns the agent loop, the tools, the planner, the workers, the task graph and the session history.

It is for pentesters, red teams, bug bounty hunters and security engineers who work from a terminal. It does the in-scope work and does not refuse it. The Acceptable Use Policy (https://excessionlabs.ai/aup/) sets the scope: you test only systems that you own or have written permission to test.

## How it works

- **A planner splits the job.** You give a goal. The planner breaks it into intents, which are tasks in a task graph.
- **Workers run the intents in parallel.** The default limit is 12 workers. Change it with `/workers`, see worker activity with `/agents`, and pause or resume with `/control`. The composer stays open while work runs.
- **You set the rules for each tool.** Each command, file write and network call can run on its own, be blocked, or wait for your approval. `excession rules` manages the rules. `--sandbox read-only` and `--sandbox workspace-write` add containment.
- **Findings come with proof.** A finding has a severity, a status, the target and source line, a summary, the evidence, steps to reproduce and a fix. Retest a finding in a fresh session. `/report` renders the findings with their recorded evidence.

Excession works with the tools you already use: MCP servers, skills, `AGENTS.md` project instructions, custom slash commands and hooks. Long sessions compact their own context. You can save, resume, fork and export sessions.

## Models

Run `/model` to choose one of two models. Both run through your Excession account. You do not need separate provider API keys.

- `deepseek/deepseek-v4.1-flash-1017` (the default)
- `deepseek-v4.1-flash-abliterated`

`/effort` sets the reasoning level. The default is high.

## How to start

1. Install the CLI. On macOS and Linux:

   ```sh
   curl -fsSL https://excessionlabs.ai/install.sh | sh
   ```

   On Windows (PowerShell):

   ```powershell
   irm https://excessionlabs.ai/install.ps1 | iex
   ```

2. Sign in. Run `excession` or `excession login`. The terminal shows a code and opens the browser. Make sure that the codes match, then approve. There is no access request and no review.
3. Add funds from $10 or pick a plan in Billing (https://excessionlabs.ai/app/billing). See how pricing works: https://excessionlabs.ai/blog/simple-pricing/
4. Open your project and give a concrete first request:

   ```sh
   excession whoami
   excession -C /path/to/your/project
   ```

   ```text
   Review this repository's authentication flow. Explain confirmed issues with file
   references and reproduction steps, then propose fixes and regression tests.
   ```

## Scripts and CI

Headless mode runs without the interface. With `--json`, stdout is a JSONL event stream, and the exit code tells the result.

```sh
excession exec 'Review this repository and verify confirmed findings.' --json
git diff | excession exec 'review this'
```

## The API

The model behind the agent is also an API. The base URL is `https://api.excessionlabs.ai/v1`. It accepts OpenAI Chat Completions, OpenAI Responses and Anthropic Messages, with streaming and tool calls. The limit is 600 requests per minute per key. The CLI and your API keys use the same account, plan credit and balance. Read the API page (https://excessionlabs.ai/api.md) and the API docs (https://excessionlabs.ai/docs.md).