SIGN IN SIGN UP
CherryHQ / cherry-studio UNCLAIMED

AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs

0 0 129 TypeScript

feat(agents): add CherryClaw autonomous agent type (#13359)

### What this PR does

Before this PR:
Cherry Studio only supports `claude-code` as an agent type. Agents have
no autonomous scheduling, no IM channel integration, and no
soul/personality system.

After this PR:
Introduces **CherryClaw** — a new autonomous agent type with:
- **Soul-driven personality**: Markdown-based soul files with
mtime-cached reading
- **Task-based scheduler**: Poll-loop scheduler with drift-resistant
interval computation, tasks as first-class DB entities
(nanoclaw-inspired)
- **Internal claw MCP server**: `cron` tool (add/list/remove)
auto-injected into CherryClaw sessions for autonomous task management
- **Channel abstraction layer**: Pluggable adapter pattern with Telegram
as the first implementation (grammY, long polling, streaming drafts,
typing indicators)
- **Headless message persistence**: Channel and scheduler messages now
persist to the agent SQLite DB
- **Basic sandbox mode**: PreToolUse hook path enforcement + OS-level
sandbox toggle
- **Full UI**: Agent creation modal with type selector, settings tabs
(soul, tasks, channels, advanced), task management CRUD, channel catalog
with inline config
- **53 unit tests** across 8 test files covering all new services

<!-- Fixes # -->

### Why we need it and why it was done in this way

CherryClaw enables Cherry Studio agents to operate autonomously —
executing scheduled tasks and responding to IM messages without user
interaction. This is the foundation for "always-on" AI assistants.

The following tradeoffs were made:
- **Poll-loop scheduler over timer-based**: DB is the source of truth;
no timer state to restore on restart. Simpler, more robust at the cost
of up to 60s latency.
- **AgentServiceRegistry pattern**: Replaced hardcoded
`ClaudeCodeService` in `SessionMessageService` with a registry mapping
`AgentType` → service. Extensible for future agent types.
- **Internal MCP server for cron**: Rather than extending the SDK's tool
system, the `cron` tool is served as a standard MCP server at
`/v1/claw/:agentId/claw-mcp`. This lets the agent discover and use it
naturally.
- **Channel abstraction over direct Telegram integration**:
`ChannelAdapter` + factory registration enables future Discord/Slack
adapters without touching core routing logic.
- **Basic sandbox (not security boundary)**: PreToolUse hook + OS
sandbox provides best-effort restriction for well-behaved agents. Known
bypass vectors documented; hardening deferred.

The following alternatives were considered:
- cron-based OS scheduling (rejected: harder to manage lifecycle, no DB
integration)
- Direct Telegram bot API calls (rejected: grammY provides typed API,
connection management, and middleware)
- Modifying SDK builtin tools (rejected: internal MCP server is cleaner
separation)

### Breaking changes

None. This is a new agent type (`cherry-claw`) alongside the existing
`claude-code` type. No existing behavior is modified.

### Special notes for your reviewer

- **New DB migration**: `0003_wise_meltdown.sql` adds `scheduled_tasks`
and `task_run_logs` tables (agents DB only, not IndexedDB)
- **New dependencies**: `cron-parser` ^5.5.0, `grammy` ^1.41
- **Placeholder avatar**: `cherry-claw.png` is currently a copy of
`claude.png` — needs a proper distinct image
- **74 files changed, ~7400 lines added** — large PR, recommend
reviewing by phase (type system → backend services → MCP → channels → UI
→ tests)
- **Sandbox is basic only**: The PreToolUse path checking has known
bypasses (relative paths, variable expansion). Documented in handoff.md.
Hardening is follow-up work.
- The `handoff.md` file in the repo root contains full architectural
context and decisions

### Checklist

- [x] PR: The PR description is expressive enough and will help future
contributors
- [x] Code: [Write code that humans can
understand](https://en.wikiquote.org/wiki/Martin_Fowler#code-for-humans)
and [Keep it simple](https://en.wikipedia.org/wiki/KISS_principle)
- [x] Refactor: You have [left the code cleaner than you found it (Boy
Scout
Rule)](https://learning.oreilly.com/library/view/97-things-every/9780596809515/ch08.html)
- [ ] Upgrade: Impact of this change on upgrade flows was considered and
addressed if required
- [ ] Documentation: A [user-guide update](https://docs.cherry-ai.com)
was considered and is present (link) or not required. Check this only
when the PR introduces or changes a user-facing feature or behavior.
- [ ] Self-review: I have reviewed my own code (e.g., via
[`/gh-pr-review`](/.claude/skills/gh-pr-review/SKILL.md), `gh pr diff`,
or GitHub UI) before requesting review from others

### Release note

```release-note
New CherryClaw agent type: autonomous agents with soul-driven personality, task-based scheduling (cron/interval/one-time), internal cron MCP tool for self-managed tasks, Telegram channel integration with streaming responses, and basic sandbox mode for filesystem restriction.
```

---------

Signed-off-by: Vaayne <liu.vaayne@gmail.com>
Signed-off-by: suyao <sy20010504@gmail.com>
Signed-off-by: Siin Xu <31815270+SiinXu@users.noreply.github.com>
Signed-off-by: zhangjiadi225 <625013594@qq.com>
Signed-off-by: greycheng255 <greycheng255@gmail.com>
Co-authored-by: kangfenmao <kangfenmao@qq.com>
Co-authored-by: suyao <sy20010504@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Siin Xu <31815270+SiinXu@users.noreply.github.com>
Co-authored-by: fullex <106392080+0xfullex@users.noreply.github.com>
Co-authored-by: zhangjiadi225 <625013594@qq.com>
L
LiuVaayne committed
a4f5191e425237dcecde5e071a624c7aa3eb25f1
Parent: 4083bd7
Committed by kangfenmao <kangfenmao@qq.com> on 4/3/2026, 6:18:02 AM