@magnitudedev/cli 0.0.1-alpha.5 → 0.0.1-alpha.7
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +35 -51
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -2,13 +2,19 @@
|
|
|
2
2
|
<img src="wordmark.svg" alt="Magnitude" width="100%" />
|
|
3
3
|
</p>
|
|
4
4
|
|
|
5
|
-
<p align="center"><b>Open source coding agent</b></p>
|
|
6
|
-
|
|
7
5
|
<p align="center">
|
|
8
|
-
<a href="https://docs.magnitude.dev" target="_blank"><img src="https://img.shields.io/badge/📕-Docs-232f41?style=flat-square&labelColor=
|
|
6
|
+
<a href="https://docs.magnitude.dev" target="_blank"><img src="https://img.shields.io/badge/📕-Docs-232f41?style=flat-square&labelColor=0369a1&color=gray" alt="Documentation" /></a> <img src="https://img.shields.io/badge/License-Apache%202.0-232f41?style=flat-square&labelColor=0369a1&color=gray" alt="License" /> <a href="https://discord.gg/VcdpMh9tTy" target="_blank"><img src="https://img.shields.io/discord/1305570963206836295?style=flat-square&logo=discord&logoColor=white&label=Discord&labelColor=5865F2&color=gray" alt="Discord" /></a> <a href="https://x.com/usemagnitude" target="_blank"><img src="https://img.shields.io/badge/-Follow%20Magnitude-000000?style=flat-square&labelColor=000000&color=gray&logo=x&logoColor=white" alt="Follow Magnitude" /></a>
|
|
9
7
|
</p>
|
|
10
8
|
|
|
11
|
-
|
|
9
|
+
|
|
10
|
+
Magnitude is an **open source coding agent** that orchestrates a team of subagents, each specialized for a role in the development process. A lead agent breaks down your task and delegates to explorers, planners, builders, reviewers, debuggers, and a browser agent.
|
|
11
|
+
|
|
12
|
+
- **Everything runs in subagents, including implementation.** The lead's context stays clean, so it remains coherent across long tasks.
|
|
13
|
+
- **The agent follows a process without you enforcing it.** No praying the agent invokes your skills. No manually triggering them when it doesn't.
|
|
14
|
+
- **A lead agent manages the work, not you.** You don't babysit, assign tasks, or decide what happens next. Go get a coffee.
|
|
15
|
+
- **Subagents stay alive.** The lead can tell the planner to consider more angles. The same builder can fix bugs that the reviewer found.
|
|
16
|
+
- **Agents share context through files, not summarization.** No output tokens burned on lossy summaries. Every handoff is a file you can view later.
|
|
17
|
+
- **Mix models by role.** Fast models for exploration. Frontier models for planning and review. You control the cost-intelligence tradeoff.
|
|
12
18
|
|
|
13
19
|
<p align="center">
|
|
14
20
|
<img src="interface.png" alt="Magnitude Interface" width="100%" />
|
|
@@ -20,7 +26,7 @@ Open source coding agent where everything happens in specialized subagents. We c
|
|
|
20
26
|
npm install -g @magnitudedev/cli
|
|
21
27
|
```
|
|
22
28
|
|
|
23
|
-
|
|
29
|
+
Then navigate to the directory you want to work in and launch the TUI:
|
|
24
30
|
|
|
25
31
|
```bash
|
|
26
32
|
magnitude
|
|
@@ -30,69 +36,47 @@ This will launch Magnitude with a setup wizard for configuring providers and mod
|
|
|
30
36
|
|
|
31
37
|
### Providers
|
|
32
38
|
|
|
33
|
-
Magnitude works with most major model providers out of the box.
|
|
34
|
-
|
|
35
|
-
**Subscription-based** — Use models you already pay for, no API key needed.
|
|
36
|
-
- **ChatGPT Plus/Pro** — Sign in with your OpenAI account
|
|
37
|
-
- **GitHub Copilot** — Sign in with your GitHub account
|
|
38
|
-
|
|
39
|
-
**API key** — Bring your own key from any of these providers.
|
|
40
|
-
- Anthropic, OpenAI, Google Gemini, OpenRouter, Vercel AI Gateway, Google Vertex AI, Amazon Bedrock, MiniMax, Z.AI, Cerebras
|
|
41
|
-
|
|
42
|
-
**Local models** — Connect to Ollama or any OpenAI-compatible endpoint. No auth required.
|
|
43
|
-
|
|
44
|
-
See the [providers docs](https://docs.magnitude.dev/configuration/providers) for setup instructions. If you would like another provider to be supported, please feel free to create an issue or raise a PR yourself.
|
|
39
|
+
Magnitude works with most major model providers out of the box, including open source and local models.
|
|
45
40
|
|
|
46
|
-
|
|
41
|
+
You can use your **ChatGPT Plus/Pro** or **GitHub Copilot** subscription.
|
|
47
42
|
|
|
48
|
-
|
|
49
|
-
- **Steerable subagents.** Subagents can surface blockers, ask clarifying questions, and be paused, resumed, or redirected mid-task.
|
|
50
|
-
- **Throw out the skills files.** Plan, review, debug, and browse are built-in subagents. No configuration needed, and the main agent actually uses them consistently.
|
|
51
|
-
- **Mix your models.** Chat with Claude, build with Codex, or use whatever combination fits your preferences for intelligence, speed, and cost.
|
|
52
|
-
- **Scoped toolsets, not modes.** Real coding jumps between planning and building too often for explicit mode toggles. Each subagent gets tools scoped to its role instead.
|
|
53
|
-
- **Talk to it while it works.** The main agent isn't blocked by implementation. Ask questions, clarify mid-build, or start planning what's next without interrupting anything.
|
|
43
|
+
See the [provider docs](https://docs.magnitude.dev/configuration/providers) for full provider support.
|
|
54
44
|
|
|
55
|
-
##
|
|
45
|
+
## How it works
|
|
56
46
|
|
|
57
|
-
|
|
58
|
-
The main agent manages the conversation and delegates to specialized subagents (explorer, planner, builder, reviewer, browser, debugger), each with its own context window, toolset, and permissions. Subagents do focused work and report back, keeping the main agent's context clean.
|
|
47
|
+
The lead agent manages all subagents on your behalf. It can message, stop, resume, or redirect them and run many in parallel. You can also message subagents directly.
|
|
59
48
|
|
|
60
|
-
|
|
61
|
-
|
|
49
|
+
Magnitude comes out of the box with the following subagents:
|
|
50
|
+
- **Explorer**: for doing codebase or web research, both broad and narrow
|
|
51
|
+
- **Planner**: for evaluating various implementation strategies
|
|
52
|
+
- **Builder**: for implementing code changes directly in your files
|
|
53
|
+
- **Reviewer**: for strict, independent review of code changes
|
|
54
|
+
- **Debugger**: for root causing bugs and fixing them
|
|
55
|
+
- **Browser**: for verifying UI changes with a built-in browser agent
|
|
62
56
|
|
|
63
|
-
|
|
64
|
-
Shared markdown documents pass context between agents directly. An explorer writes findings, a planner reads them and produces a plan, a builder reads the plan and implements. No context is lost to summarization, and the main agent doesn't burn output tokens on handoff.
|
|
57
|
+
Magnitude may use none or all of these in a given session. For a quick fix in a single file, it may edit it directly. For a very in-depth change, it may use the whole team. For most tasks, it will use some combination of explorer, planner, builder, and reviewer.
|
|
65
58
|
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
|
|
69
|
-
### Progressive disclosure of tool output
|
|
70
|
-
Tool output does not enter the context window by default. Agents inspect results explicitly and pull in only what they need, keeping noisy output out of the conversation.
|
|
59
|
+
<p align="center">
|
|
60
|
+
<img src="architecture.png" alt="Magnitude Architecture" width="100%" />
|
|
61
|
+
</p>
|
|
71
62
|
|
|
72
|
-
|
|
73
|
-
Commands are classified by risk. Read-only commands run freely. Normal commands are allowed in your project directory but blocked outside it. Dangerous commands are always blocked. No `--dangerously-skip-permissions`, no stuck permission prompts.
|
|
63
|
+
## Why we built this
|
|
74
64
|
|
|
75
|
-
|
|
76
|
-
A vision-based browser agent built on [browser-agent](https://github.com/magnitudedev/browser-agent) runs natively in the same runtime for verifying UI changes and behavior as part of the workflow.
|
|
65
|
+
The community clearly wants more out of Claude Code. Projects like Superpowers have hit 100k+ GitHub stars augmenting Claude Code with more skills/subagents. People want agents to follow better software development processes.
|
|
77
66
|
|
|
78
|
-
|
|
79
|
-
Magnitude learns your preferences and codebase conventions over time, storing them in `.magnitude/memory.md` and applying them automatically to future sessions. Skills and AGENTS.md are also supported.
|
|
67
|
+
But they're plugging into an existing coding agent that wasn't built to use them consistently and effectively. The result is inconsistent usage and mixed results, often having to repeatedly prompt the agent to actually use the skills.
|
|
80
68
|
|
|
81
|
-
|
|
69
|
+
Instead of plugging into an existing primitive, we built a new one from the ground up. Imagine Claude Code was actually built around Superpowers. **That's Magnitude.**
|
|
82
70
|
|
|
83
|
-
|
|
71
|
+
## Documentation
|
|
84
72
|
|
|
85
|
-
|
|
73
|
+
Full documentation is available at [docs.magnitude.dev](https://docs.magnitude.dev).
|
|
86
74
|
|
|
87
|
-
|
|
75
|
+
## Contributing
|
|
88
76
|
|
|
89
77
|
See the [contributing guide](https://docs.magnitude.dev/contributing) to get started.
|
|
90
78
|
|
|
91
|
-
|
|
92
|
-
|
|
93
|
-
Full documentation is available at [docs.magnitude.dev](https://docs.magnitude.dev).
|
|
94
|
-
|
|
95
|
-
### Acknowledgements
|
|
79
|
+
## Acknowledgements
|
|
96
80
|
|
|
97
81
|
Built on top of [BAML](https://boundaryml.com), [Effect](https://effect.website), and [OpenTUI](https://github.com/anomalyco/opentui).
|
|
98
82
|
|