Leo’s Build Log
PostsProjectsAbout한국어

Janus

An agentic coding IDE on a local LLM

Screens

Overview

Period
Aug 2026 – present
Team
Solo
Role
Planning · design · desktop app · backend · local inference

The problem

When a coding agent finishes answering, checking the changes, testing, reviewing and committing are still left. The starting question: how much verified work can limited local hardware produce?

The approach

  1. 01
    One Task holds the whole contextChat, terminal, editor, preview and review live in one Task, with running, needs-input, failed and awaiting-review kept apart.
  2. 02
    Local model or your existing CLIA local MLX model (Qwen 27B 4-bit) by default, or an already signed-in Claude Code or Codex CLI as the runner.
  3. 03
    Only verified changes are committedChanges are verified and reviewed against the Git diff before commit, optionally continuing to push and PR through gh.

Architecture

ArchitectureDesktopBackendAgentReviewTargetlocal HTTP · WSIPCprocess mgmtmodel serverinferencereview
React UITask · editor · terminal · preview
Electron MainProcesses · IPC
FastAPITask API · HTTP·WebSocket
SQLiteTasks · sessions
MLX RuntimeQwen 27B 4-bit
OrchestratorDelegation · budgets
ToolsFiles · shell · checks
Git diffSource of change
Verify · reviewCurrent revision
Commit · Push · PROptional gh CLI
Chosen repo · current branch
  1. 01
    Local firstOn-device inference with a local MLX model (Qwen 27B 4-bit)
  2. 02
    Resource limitsCaps on worker count and time, token and step budgets
  3. 03
    Distinct statesRun finished ≠ verified ≠ review accepted ≠ committed

User scenario

User scenarioUserSystemRetry pathapproval requestnoyes · after review
Developer
Create a TaskProject · goal
Choose a runnerLocal MLX / Claude Code · Codex
Approve toolsReview file and shell calls
Check diffOn the same screen
AcceptCommit · PR
Agent runsOrchestrator · worker
Tools runRepo's current branch
Git diff · testsCurrent revision
CommitHEAD · SHA check
Checks pass?
Find cause · rerunRun state ≠ success
  1. 01
    User flowTask → runner → tool approval → diff and verification → accept review
  2. 02
    System flowAgent run → tool execution → Git diff and tests → commit
  3. 03
    Distinct statesA finished answer ≠ a successful task — rerun when verification fails

My work

  1. 01
    Delegation with a budgetThe orchestrator hands sub-tasks to workers under caps on worker count, time, tokens and steps, with AgentProfiles deciding the tools.Orchestrator · workers · AgentProfile · resource scheduler
  2. 02
    Generation and verification overlapModel generation slots and tool/verification work are scheduled separately, so verification runs while generation waits.MLX · generation slots · queue priority · time and token caps
  3. 03
    Separate states, honestly reportedRun finished, verified, review accepted and committed/pushed are distinct states, and a push requires the recorded commit to match HEAD.Git diff · per-revision review · SHA check · gh CLI
  4. 04
    Permission boundaries per runnerThe local and Claude Code paths go through Janus's tool approvals, and each runner's limits are documented openly.MCP · default-deny approvals · HTTP/WS token and Origin checks
  5. 05
    Docs checked against codeA CI test compares key claims in the docs with the code, so old design notes never read as current features.pytest · TypeScript checks · bundle size · dependency audit · macOS packaging CI

Why these tools

로컬 LLM · MLXCode and task context must not leave the machineAgents run on on-device inference
PythonWorkers with unlimited write access and resourcesRead-only enforced; per-worker step, token and time budgets
GitUnclear origin of changes; a finished answer is not a finished taskDiff and branch define the change; verification defines done
FastAPI · SQLiteConnecting local execution to the UI and keeping Task stateLocal API plus stored Task and session records
Electron · React · TypeScriptChat, terminal, editor and review scattered across toolsChanges, verification, review and commit state on one screen

Improvements

  1. 01
    Worker efficiencyUnder the same fixed-worker policy, acceptance held at 14/15 while wall time fell from 109.9 s to 88.1 s and prompt tokens from 14,855 to 10,993 (v1 audit, 2026-08-23).
  2. 02
    Repeated runs on the real 27B model44 of 45 TaskSuite runs passed acceptance, and a 281-cycle recovery soak left zero transient state with SQLite integrity intact.
  3. 03
    Interrupted work never shows as doneAfter a restart, unfinished work is restored as interrupted, with backup/restore and a redacted diagnostics bundle.

Related posts