English · 简体中文
An open-source graphical Agent workbench that connects freely to mainstream LLM APIs, giving long, heavy coding tasks a fully visible execution environment.
MindShell is entering Source Alpha. The workbench, controlled code editing, permissions, Event Log, Checkpoints, Runtime Session rebuild, and Context Compaction are running today. The project has also been exercised with real models across long tasks and recovery scenarios.
The line above is still the product goal, not a claim that every part is finished. Broad compatibility across mainstream models has not been certified, the full long-task public test is still gated, and long-term memory, an in-house Loop, and the plugin ecosystem remain later work. The current release is source-only and limited to Apple Silicon Macs.
A complete Web-first interface: windows, tasks, steps, permissions, and processes managed on one screen. Not a shell around a command line, but a console designed for long-term use.
The Alpha is distributed as source code on GitHub. Anyone who finds the repository and wants to take part may do so; there is no invitation, application, or allowlist.
This round covers the MindShell Server, Runtime Worker, Dot Workbench Web UI, ordinary Project Binding v1, a user-configured Provider, and local-project reading, controlled editing, Permission, verification, Checkpoint, and continuation in the same Window.
It is not a Beta or stable release. There is no packaged macOS app in this round. Intel Macs, Windows, Linux, Glass UI, Web Fetch/Search, the product MCP entry point, Recall/long-term memory, and ordinary Project Authority v2 are outside the current release scope.
- Apple Silicon (arm64) Mac;
- macOS;
- Node.js 22.19.0 or newer;
- pnpm 11.0.0 or newer; the repository pins 11.7.0;
- Git;
- any local toolchain required by the project under test.
Start from a clean checkout of the Tag or Commit named in the Release:
pnpm install --frozen-lockfile
pnpm buildStart the Server and Runtime Worker in the first terminal:
pnpm --filter @mindshell/server startStart the Web UI in a second terminal:
pnpm --filter @mindshell/web devOpen http://127.0.0.1:5173.
Use a small Git repository that you can restore for the first run. Read each permission request and keep your own version-control backup.
For the first experience and reproducible issue reports, the current recommendation is DeepSeek official deepseek-v4-flash. This is test guidance, not a model allowlist. Other Providers and models may be tried, but they have not all received the same compatibility coverage.
Stop the candidate and record the Window ID if an approval works only after a refresh, the final reply appears only after a refresh, a completed task remains running, recovery repeats a settled side effect, or the same Window cannot continue.
The Source Alpha participation and release guide covers setup isolation, cold restart, diagnostics, and safe issue reporting.
Contact and feedback: Mindshell@126.com
A complete Web-first interface: windows, tasks, steps, permissions, and processes managed on one screen. Not a shell around a command line, but a console designed for long-term use.
Status: EARLY ALPHA
Find relevant code with a bounded repository map and text or symbol candidates, then bind each structured edit to freshly read file state. Before commit it checks again; if the file changed, the edit stops instead of overwriting it.
Status: EARLY ALPHA
R0–R4 risk tiers, a kernel-level Hardline, and three approval modes. Let it run; the boundary is still the one you set.
Pro mode means fewer prompts, not fewer boundaries.
Status: EARLY ALPHA
The Event Log is the single source of truth. Tool calls, permission decisions, process state, and checkpoints are recorded natively; a refresh restores from facts.
It does not rely on the wording of an answer to decide whether a step happened. Each step is recorded separately, and the UI rebuilds from the same underlying facts.
Status: EARLY ALPHA
Long-term memory stays small, high-confidence, and traceable to source; Skills are indexed on demand with a fixed origin and version.
This is a product direction. Memory Working-Set and Skill foundations have implementation and offline gates, but Recall/long-term memory is not enabled in this Alpha. The project is still working through how to judge useful, wrong, or irrelevant recall and how recalled material should be handled later.
Status: ROADMAP
Task Lease, Checkpoints, Runtime Session rebuild, and Context Compaction form a working foundation. Recovery reconciles current state before continuing on a fresh session and lease; the complete long-task public test remains gated.
A Checkpoint preserves the facts needed for recovery; it is not a workspace snapshot. Context Compaction turns older events into a traceable summary while retaining the recent exact tail.
Status: EARLY ALPHA
flowchart TD
UI["Dot Workbench"] --> API["Server / Local API"]
API --> PROJECTION["Projection"]
API --> STORE[("Event Store")]
PROJECTION --> STORE
STORE --> WORKER["Runtime Worker"]
WORKER --> KERNEL["Kernel / Tool Runner"]
KERNEL --> PROVIDER["Provider API"]
KERNEL --> PROJECT["Local Project"]
KERNEL --> PERMISSION["Permission / Guard"]
KERNEL --> CHECKPOINT["Checkpoint / Recovery"]
The Event Store is the account-scoped single writer for durable facts, sequence, Task lease, Checkpoint pointer, and the Projection outbox. Providers, Pi, Skills, plugins, and UI adapters do not receive direct authority over files, processes, networks, secrets, Permission, leases, or Event Store internals.
Pi Agent's lightweight Loop is currently a replaceable internal component. MindShell owns Provider Transport, Turn, Context, Tool, Event, and recovery, so replacing or building a new Loop later does not require replacing the facts and safety foundation.
MindShell first establishes whether an Agent can be trusted, controlled, and kept working over the long run, then expands what it can do. The roadmap starts with traceable facts and moves through the long-task public test, memory, an in-house Loop, and a plugin ecosystem toward a customisable Agent platform for accountable work.
This is a direction, not a list of features available today. The status under each heading describes its current stage.
Foundation in place
MindShell treats the Event Log as the single source of truth inside an account. Model requests, tool calls, permission decisions, process-state transitions, file changes, checkpoints, errors, and recovery actions become replayable facts rather than something inferred afterwards from the wording of an answer.
This lets the interface be rebuilt and recovery proceed without relying on the model's account of events. The same identity, causality, and execution boundaries can remain in place as the system grows.
Current mainline
The current work moves long tasks from “they run” to “they stay accurate, bounded, and recoverable once the history grows.” That means paginated, rebuildable Task, Turn, and message views; removing unbounded Event reads; and closing the working-set boundaries around large Task handoff, Project authority, and recovery.
Current macOS evidence covers real Providers, multi-file programming, 60+ Turns, and crash recovery. Equivalent Windows gates for process handling, paths, credentials, and the browser are not complete, and this Source Alpha makes no cross-platform claim.
Next phase
MindShell's memory is not meant to push the whole history back into the context. The raw conversation archive remains the source of truth, while recall brings back a small amount of relevant, high-confidence material when it is needed.
Progressive Recall is intended to load the lightweight index first, then a Memory Card, and only then the original excerpt. Ranking, removal of wrong or irrelevant recall, and the later handling of recalled material are still being designed. Thin evidence must not silently turn a new memory into fact.
The value of memory is not in remembering everything, but in recalling what matters accurately.
After the memory system
MindShell currently runs Pi Agent's lightweight Loop, but it is a replaceable internal component. Providers, context, tools, permissions, Events, Checkpoints, recovery, and product state are held by MindShell itself, so replacing the Loop later does not mean tearing down the kernel.
After the memory system and the current long-task and cross-platform gates close, MindShell plans to work on its own Loop. The aim is not simply to repeat the model-tool-model cycle, but to explore long-range planning, parallel work, recovery, reflection, verification, and tasks that cross domains.
The Loop can be replaced. The facts and safety boundaries are not replaced with it.
Planned · Ecosystem
MindShell plans one governed system for plugins, Skills, and MCP capabilities. Each capability records its origin, version, scope, and permission boundary, and continues through the same Tool Runner, Permission, Guard, and Event Log.
The default context carries only a lightweight capability index. When a task needs one, the matching plugin, Skill body, and resources can load progressively. A long task pins the versions it uses, so recovery does not silently upgrade them.
Capability can extend. Permission cannot spread.
Planned · Platform
MindShell plans to extend the Event-first kernel into an Agent platform that can be tailored to a team, project, or business process. Each Agent may have its own tools, rules, permissions, Providers, and workflow while sharing the same boundaries for facts, identity, and execution.
The Event Log is more than a runtime log; it is the ledger for the chain of responsibility. Projections can be rebuilt, and crash recovery starts from durable Events, Checkpoints, the Task Lease, and reconciliation with the current workspace. Unknown state pauses for human handling.
The Agent can be customised. Its chain of responsibility cannot be customised away.
Planned · Vertical Loop proof
MindShell plans to use a closed-source derivative for automated editing and content packaging to test the replaceable-Loop architecture in another field. It would keep MindShell's Event-first kernel, memory, permission and tool governance, Checkpoints, and crash recovery while replacing the Loop closest to task reasoning and execution.
The Loop would be specialised for relationships among source materials, script and timeline state, multi-stage edits, media tools, review feedback, and convergence across versions.
The foundation stays general; the Loop and product become specialised.
Once the in-house Loop, memory system, and plugin system form a complete path, MindShell plans to explore Multi-Agent/DAG execution, long-range planning with dynamic re-planning, cross-domain tool collaboration, automatic verification, and a broader entry point for personal work.
Coding remains the strictest and most useful place to prove reliability, but it is not intended to be the endpoint. The longer-term goal is an open Agent system that can carry complex work for individuals and teams.
MindShell is licensed under the Apache License 2.0. Third-party notices are listed in THIRD_PARTY_NOTICES.md and NOTICE.