Documentation Index
Fetch the complete documentation index at: https://docs.magnitude.dev/llms.txt
Use this file to discover all available pages before exploring further.
Magnitude is a curated system of specialized agents, each with its own defined role. Every agent is a tuned bundle of system prompt, scoped context, dedicated toolset, and a model + reasoning level.
Why specialized agents
Open models are less “one size fits all” than closed ones. Different open models are better at different things: some are fast and great at exploration, some are strong at planning, some are tasteful with code. A single model trying to do everything leaves a lot on the table.
Specialized agents let us match each role to the model that actually performs best for it. We benchmark combinations, tune prompts to each model’s quirks, and assign work accordingly. The result is a team that punches above what any single model could do on its own.
We test these constantly. New models drop, the lineup updates.
The leader
You only ever talk to the leader. The leader runs the conversation, holds your intent, and delegates work to the specialists when it needs to. You don’t pick agents or route tasks. The leader figures out who should do what.
For simple questions and small fixes, the leader handles them directly. For larger work, it delegates. When a specialist finishes, the leader reviews the output and decides what’s next.
Model: Kimi K2.6
The specialists
Seven specialists the leader delegates to. Each has its own role, model, and tool access.
Scout
Fast and efficient exploration. Scouts investigate the codebase and gather evidence quickly. Read-only access.
Model: MiniMax M2.7
Architect
Plans and high-level design thinking. Architects evaluate approaches and produce implementation plans before code gets written.
Model: Kimi K2.6
Engineer
Concrete planning and implementation. Engineers do the actual code changes.
Model: DeepSeek V4 Pro
Critic
Critical and detail-oriented analysis. Critics independently verify quality and behavior after changes.
Model: GLM 5.1
Scientist
Empirical debugging and information gathering. Scientists root-cause bugs through hypothesis testing. The focus is diagnosis, not implementing the fix.
Model: DeepSeek V4 Pro
Artisan
Tasteful and creative work. Artisans handle non-code deliverables like docs, configuration, and content.
Model: Kimi K2.6
Advisor
Smart peer of the leader, always available. Advisors give the leader a second opinion when something is uncertain or high-stakes.
Model: GLM 5.1
