magnitude - Moses Yebei

The Problem

Most AI coding agents depend on cloud APIs, which means sending sensitive code and data to third parties, paying per-token costs, and dealing with rate limits. Teams that handle private or regulated data need agentic coding assistance that runs entirely on their own hardware, with no network egress.

What This Does

Magnitude is an open-source agent that runs fully local—models, inference, and tool execution—with zero cloud dependencies. It packages a custom Rust inference engine built on llama.cpp (in inference/), a TypeScript agent core (in packages/agent/), and a CLI (in cli/) that profiles hardware, recommends models, and manages loading/switching.

The agent can inspect/edit files, run shell commands, and manage long sessions. Skills extend it to handle Excel, PowerPoint, PDFs, and browser automation. The packages/ directory holds the bulk of the logic (1,257 files), with inference/ as the Rust engine, cli/ as the entry point, and web//desktop/ as UI shells.

How It Is Wired

Execution starts at cli/src/index.tsx, which boots the interactive terminal UI. From there, the app routes through cli/src/features/local-inference/index.ts to the agent core in packages/agent/, and to the Rust engine via inference/crates/icn-api/. The agent's events module is the central hub—98 modules depend on it, making it the highest-blast-radius file in the codebase.

The agent lifecycle (packages/agent/src/projections/agent-lifecycle.ts) tracks state transitions and sits in a circular dependency with events.ts and turn.ts. That cycle means a change to any of those three files can ripple unpredictably through the import graph. The packages/acn/src/agent-runtime.ts file handles runtime orchestration and is flagged for deep nesting (indentation depth 6), which makes control flow hard to follow.

The inference engine (inference/) is a separate Rust workspace with crates for API, catalog, and benchmarking. It handles memory calculation, acceleration tuning, and batching before model loading. The packages/acn/ directory appears to handle agent-communication-network logic, though its wiring to the CLI is not fully mapped in this analysis.

How To Use It

npm install -g @magnitudedev/cli
cd your-project
magnitude

The CLI supports macOS and Linux (Windows via WSL). No API keys or environment variables are required. On first run, it profiles hardware and prompts you to pick a model tier (Balanced, Best Quality, Fastest, Lightweight), then handles download and configuration. Skills are added via npx skills add <repo>.

Real-World Use

A financial analyst working with confidential client data runs magnitude inside a secured VM. They use it to review code for PII leaks, generate audit summaries, and build Excel reports via the xlsx skill—all without data leaving the machine. The built-in inference engine keeps context windows intact across parallel agent sessions, so multi-file refactors remain coherent.

Code Health & Issues

Static analysis (not opinion) found 288 findings: 99 high, 188 medium, 1 low. Key items:

High – Import cycle: packages/agent/src/events.ts, projections/agent-lifecycle.ts, and projections/turn.ts are mutually reachable. Breaking this requires extracting shared types or deferring imports.
High – Oversized files: events.ts (602 lines), agent-lifecycle.ts, and acn/src/agent-runtime.ts exceed maintainable size.
High – Hub modules: events.ts has 98 dependents; churn there is high-blast-radius.
Medium – High branching density in cli/src/components/button.tsx, cli/src/utils/clipboard.ts, and acn-protocol/src/acn-identity.ts.

SDLC observations: CI exists but the 4 workflows read do not invoke the 339 test files—a green check that never ran assertions. Third-party GitHub Actions are pinned to tags (oven-sh/setup-bun@v2) rather than commit SHAs. No Dependabot/Renovate config. Two large binaries (9.0MB GIF, 5.6MB MP4) bloat every clone. No dependency vulnerability scan in CI.

The Bottom Line

Magnitude is a serious, well-structured local-agent project with real engineering depth—the Rust inference engine and the agent core are substantial. The circular dependency and hub modules in the agent layer will make changes there costly. Teams needing fully offline, private agentic coding should evaluate it; teams comfortable with cloud APIs have cheaper options.

What the analyser found

Deployment readiness

6/7

Container image
CI pipeline
Lockfile committed
Test suite
README
License
No committed secrets

Composition

1,862 files

TypeScript1290
TSX142
Markdown122
JSON116
Rust72
TOML38

ReactSvelteLaravelTransformers

Architecture

Top-level areas of the codebase, sized by module count. Arrows show how many imports cross from one area into another.

graph TD g0["packages 980 modules"] g1["cli 161 modules"] g2["web 37 modules"] g3["inference 14 modules"] g4["desktop 6 modules"] g5["scripts 2 modules"] g3 -->|2| g5 g5 -->|1| g3

Module dependencies

The 10 most depended-upon modules of 1,421, from static import analysis. Red outline marks a module in an import cycle.

graph LR n0_packages_agent_src_events["events"] n1_cli_src_hooks_use_theme["use-theme"] n2__projections_agent_lifecycle["agent-lifecycle"] n3_s_agent_src_projections_turn["turn"] n4_ions_user_message_resolution["user-message-resolution"] n5__agent_src_models_tool_state["tool-state"] n6_s_agent_src_projections_goal["goal"] n7_rc_projections_agent_routing["agent-routing"] n8_kages_agent_src_window_index["index"] n9_rc_projections_harness_state["harness-state"] n6_s_agent_src_projections_goal --> n0_packages_agent_src_events n4_ions_user_message_resolution --> n0_packages_agent_src_events n9_rc_projections_harness_state --> n0_packages_agent_src_events n7_rc_projections_agent_routing --> n0_packages_agent_src_events n3_s_agent_src_projections_turn --> n0_packages_agent_src_events n3_s_agent_src_projections_turn --> n2__projections_agent_lifecycle n3_s_agent_src_projections_turn --> n7_rc_projections_agent_routing n3_s_agent_src_projections_turn --> n4_ions_user_message_resolution n3_s_agent_src_projections_turn --> n6_s_agent_src_projections_goal n2__projections_agent_lifecycle --> n0_packages_agent_src_events classDef cyc stroke:#ef4444,stroke-width:2px; class n0_packages_agent_src_events,n2__projections_agent_lifecycle,n3_s_agent_src_projections_turn,n4_ions_user_message_resolution,n6_s_agent_src_projections_goal,n7_rc_projections_agent_routing,n8_kages_agent_src_window_index,n9_rc_projections_harness_state cyc;

What would stop me shipping this

Ranked by severity × confidence × production reach. Reach is the honest discriminator across a collection that is mostly other people's code: the same finding matters more in something that ships.

high2

medium4

low2

Pin third-party GitHub Actions to a commit SHAhigh2 occurrences

.github/workflows

oven-sh/setup-bun@v2, dtolnay/rust-toolchain@stable

A tag can be moved, so the action running with your token and secrets is whatever its owner last pushed; this is how tj-actions/changed-files leaked secrets from thousands of repos.

Fix: Replace each @vN with the 40-character commit SHA, keep # vN as a comment, and let Dependabot bump the SHAs.

Make CI invoke the test suite it hashigh339 occurrences

.github/workflows

339 test files, no test command in the 4 workflows read

A green check that never executed an assertion is worse than no check, because reviewers trust it.

Fix: Add a test step to the existing workflow rather than a new workflow.

Enable Dependabot or Renovatemedium

50 manifest(s), no update bot configured

Without a bot a published advisory sits unpatched until someone audits by hand, which across 1,322 repositories means never.

Fix: Commit .github/dependabot.yml covering the repo ecosystems plus github-actions.

Gate pull requests on a dependency vulnerability scanmedium

.github/workflows

no dependency scan in CI

This is the one gate that would catch a known-vulnerable package before it reaches a build, and no repository in the sample had it.

Fix: Add dependency-review-action on pull_request, or osv-scanner on push and a schedule.

Move large binaries to Git LFS or out of the repositorymedium2 occurrences

docs/maglocaldemo.gif

2 blobs over 5MB: maglocaldemo.gif 9.0MB, maglocaldemo.mp4 5.6MB

One repository here carries twenty blobs over 5MB including a 10.9MB spreadsheet, so every clone and every CI checkout pays for data nobody diffs.

Fix: Track those extensions with LFS, or move datasets to object storage and fetch them in a setup step.

Set persist-credentials: false on checkoutmedium

.github/workflows/publish-npm.yml

checkout keeps the token, then dependencies are installed

The token stays in .git/config for every later step, so a malicious postinstall script reads a pushable credential without one ever being passed to it.

Fix: Add with: persist-credentials: false, and pass an explicit token only to the step that pushes.

Set timeout-minutes on the workflow jobslow4 occurrences

.github/workflows/publish-npm.yml

4 workflow(s) declare no job timeout

A wedged step runs to the six-hour platform default, which on a two-hourly schedule means three runs overlap behind it.

Fix: Add timeout-minutes with a realistic bound to each job.

Add the repository convention files this project lackslow3 occurrences

missing .editorconfig, .gitattributes, a formatter config

Without them one contributor's editor writes tabs into a Python file, a shell script commits with CRLF and fails in the container, and a notebook diff is unreviewable.

Fix: Add .editorconfig, .gitattributes with text=auto eol=lf, and a formatter config.

Checked deterministically against the repository tree and a bounded set of its files: committed credentials, unpinned actions and base images, missing lockfiles and update bots, workflows that discard failures, published advisories against the declared dependencies, runtime configuration, licensing and notebook reproducibility. No language model is involved in this section.