Brigade: A Self-Hosted Multi-Agent Crew with Persistent Memory and No SaaS Middleman

What Brigade Actually Solves

Brigade addresses a specific pain: running multiple AI agents that coordinate like a team, while keeping all data and credentials on your own machine. Most multi-agent frameworks either force you into a hosted platform or leave you stitching together memory, tool access, and inter-agent communication yourself. Brigade bundles these into a single CLI tool that creates a crew of agents with distinct personas, workspaces, and credentials. The documentation frames it as "your personal intelligence, built enterprise-grade," which means the design targets users who want the control of self-hosting without giving up the convenience of a ready-made system. The intended audience appears to be developers and technical operators who are comfortable with a terminal-first workflow and who need to automate tasks across many applications without sending API keys to a third party.

The Org Chart and Tideline Memory Mechanism

The core architectural idea in Brigade is that agents exist in a "real org chart" that governs delegation and permissions. This is not a flat list of bots; it is a hierarchy where agents have roles and can delegate tasks to each other. The README states that agents "share one long-term memory, Tideline," so that what one agent learns becomes available to others. Tideline is described as a memory engine with "origin scoping, decay, and hybrid (keyword + vector) recall." Origin scoping means memory entries are tied to the agent or context that created them, which prevents irrelevant information from leaking across the crew. Decay likely reduces the influence of old or unused memories over time. The hybrid recall combines keyword search with vector embeddings, which is a common pattern for making memory both precise and semantically flexible. This mechanism is central to Brigade's value proposition: without Tideline, each agent would start fresh every session, and the crew would lose the benefit of accumulated knowledge.

Installation and First Run: Real Commands

Getting Brigade running is a single shell command on macOS or Linux. The README provides: `curl -fsSL https://brigade.spinabot.com/install.sh | sh` followed by `brigade`. The installer is said to install Node if needed, and no sudo is required. After starting `brigade`, you get a chat TUI. The project is distributed as an npm package, `@spinabot/brigade`, which suggests that manual installation via `npm install -g` is also possible, though the README only shows the curl method. Configuration lives under a single `~/.brigade/` directory, which the user fully owns. There is no account creation and no setup step mentioned, aside from choosing a model provider. You can bring your own API key for Claude, GPT, Gemini, or Llama, or use a local Ollama instance. The README also mentions that you can reuse an existing Claude Code or Codex CLI login, avoiding browser re-authentication. This is a practical touch for users who already work with those CLIs.

Storage Modes: Filesystem vs. Convex

Brigade offers two storage backends. The default is a small filesystem install under the `~/.brigade/` directory. This is simple and keeps everything local. The alternative is a self-hosted Convex database, which the README says you can switch to "when you want one." Convex is a backend platform with a reactive database, and self-hosting it means running your own Convex server. The choice affects scalability and concurrency: a filesystem is fine for a single user or a small crew, but if you have many agents writing memory concurrently, a database becomes necessary. The README does not provide the exact command or config key to switch storage modes, which is a gap for anyone trying to set this up. The documentation says "switch to a self-hosted Convex database," but without a config example, you would need to dig into the source or wait for the docs to fill in. This is a point where the marketing outpaces the practical guide.

The 'Bloody Benchmark' and Exposure Feature

A distinctive feature is the `brigade bloody benchmark` command, introduced in v1.9.0. The README describes it with theatrical language: it "throws your crew to the wolves" by exposing your agents to the open internet. The underlying mechanism is an HTTPS tunnel, using Cloudflare by default, or a bring-your-own relay like `bore`, `frp`, or `sish`. A secret key is attached invisibly to requests, and unauthorized access is met with a `401`. The command `brigade expose` is the more formal name for this feature, and `brigade expose stop` ends the exposure. This is a real testing tool: instead of benchmarking agents in a lab, you let real strangers interact with them. It is a clever idea, but it carries obvious security risks. The README claims gate-crashers "bleed out on a 401," but any exposure of an agent to the public internet is a potential attack surface. The secret key must be strong, and the agent must be sandboxed. The documentation does not detail what happens if an attacker bypasses the 401 or what protections exist beyond the key.

Limitations and Wrong-Tool Cases

Brigade is not for everyone. The terminal-first design means there is no polished GUI; the README boasts "no browser, no Electron," which is a feature for some but a barrier for others. Non-technical users who want a visual dashboard will find the TUI insufficient. The project also has a steep feature list: org charts, skill systems, cron schedulers, sub-agent fan-out, and 1,000+ app connectors. That breadth can be overwhelming, and the README does not provide a clear onboarding path beyond the basic chat. The documentation is thin in several areas: there is no explicit list of configuration keys, no troubleshooting guide, and no detailed explanation of how the org chart is defined. The release cadence is rapid (v1.37.1 in September 2026, with multiple releases per week), which suggests active development but also means APIs and config may change frequently. If you need a stable, long-term platform with minimal maintenance, this pace could be a liability. Brigade is also wrong if you require a fully managed service: the project explicitly avoids a SaaS middleman, so you are responsible for updates, security, and your own infrastructure.

Alternatives and Their Different Approaches

The most direct alternative is a hosted multi-agent platform like CrewAI or AutoGen, but those typically require you to run your own orchestration code and often rely on cloud APIs. A closer comparison is something like LangChain's agent framework, which gives you building blocks but not a ready-made crew with memory. The key difference with Brigade is the turnkey nature: it is a single binary-like CLI that sets up a crew, memory, and connectors out of the box. Another alternative is to use a single-agent CLI like Claude Code or Codex on its own, which is simpler but lacks multi-agent coordination and shared memory. If you need multi-agent orchestration with a GUI, you might look at n8n or Flowise, but those are workflow tools rather than autonomous agent crews. Brigade's approach is to give you a complete system that you self-host, which is a middle ground between building everything yourself and paying for a hosted service. The trade-off is that you must trust the project's design decisions and stay on top of its evolution.

Maintenance, License, and Upgrade Considerations

Brigade is MIT-licensed, which means you can use, modify, and distribute it freely, including in commercial products, as long as you preserve the copyright notice. There is no copyleft obligation, which is a plus for enterprises that want to embed it. The project is written in TypeScript and runs on Node, with the npm badge indicating a supported Node version. Maintenance costs are tied to the release cadence: the repository shows three releases within nine days in September 2026, which indicates a fast-moving codebase. You will need to update regularly to get bug fixes and security patches, but each update may introduce breaking changes. The README does not mention a migration guide or a changelog, so you should check the release notes on GitHub before upgrading. The project is not archived, and the last push is recent, so it is actively maintained. However, the documentation is incomplete in places, so you may need to read the source code to understand certain features. The self-hosted Convex option adds its own maintenance burden, as you are running a database server. Overall, Brigade is a young, ambitious project with a strong vision, but you should verify its maturity against your own requirements.

Editorial conclusion

Adopt Brigade if you are a developer or small team that wants a self-hosted, multi-agent system with persistent memory, no mandatory SaaS, and the flexibility to use Claude, GPT, or local Ollama models. Skip it if you need a polished, GUI-first product or if you cannot invest time in CLI-based configuration and troubleshooting. Before adopting, verify the current state of the Convex backend integration, test the agent isolation features with your own workloads, and confirm that the 1,000+ app connectors cover the specific tools you rely on. The project is under active development, so check the latest release notes and the GitHub issues for any breaking changes.

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. spinabot/brigade on GitHub