PicoClaw Roadmap: March 2026 (Week 2)

[Zepan]

🖥️ 1. WebUI Enhancements

Current pain point: Many users get stuck during the initial configuration. The UI needs to be more intuitive and secure.

• Frontend Optimization: Further optimize the web frontend pages for a smoother UX.
• Security Interface: Add a dedicated UI panel for security and permission configurations.
• Docker WebUI Support: Enable WebUI access within Docker deployments (Must include strict security/authentication measures).
• Configuration Wizard: Implement a setup wizard to guide new users through the initial configuration process.

🧠 2. Agent Refactor (Kick-off)

Aligning with the ongoing Agent model consolidation track.

• Agent Loop & Events: Implement a robust Agent Loop and event notification system.
• Intermediate Outputs: Stream intermediate results during tool execution loops to prevent the UX from feeling "stuck" or unresponsive.
• Execution Interruption: Allow users to manually interrupt/cancel an ongoing tool execution loop.
• Memory System: Refactor context and memory management.

🔒 3. Security

• Credential Isolation: Extract sensitive information (e.g., LLM API keys, secrets) from config.json (e.g., migrate to .env or a secure keystore).

👁️ 4. Multimodal LLM Support

• Provider Config Updates: Add modality tags (e.g., vision, audio) and an enable toggle flag in the provider configurations.
• API Integration: Complete API integrations for speech-to-text (audio) and image recognition (vision) across mainstream multimodal models.

💬 5. Channels

• Multimodal Attachments: Continue improving parsing and handling of multimodal attachments across all supported messaging channels.

🤹 6. Skills & Tools

• ClawHub: Add configuration support for the ClawHub token.
• Search Tool: Implement/Enhance the dedicated Search tool.

🔌 7. MCP (Model Context Protocol) Integration

• mobile-use MCP.
• browser-use MCP.
• computer-use MCP.

📱 8. Android Support

• Termux Guide: Publish a comprehensive, step-by-step tutorial for deploying PicoClaw on Android via Termux.
• Native APK: Develop and release a native Android APK for PicoClaw.

🤖 9. AI CI & Automation

• Stale Bot: Implement a bot to automatically close stale/inactive Issues and PRs.
• Smart PR Management: Set up a cron job to close old conflicting PRs with no author response. For highly valuable PRs, use an AI bot to resolve conflicts and recreate the PR automatically.
• Release Testing: Add comprehensive automated test cases for the release pipeline.

🌐 10. Documentation, Website & Community

• Official Website: Beautify the homepage UI and optimize SEO.
• Docs Site: Build and deploy a dedicated documentation website (e.g., VitePress/Docusaurus).
• Repo Beautification: Improve the GitHub repository appearance (README, banners, feature badges).
• Showcase: Create and publish high-quality promotional showcase examples and use cases.

[nikolasdehor]

A few notes on the current state of the codebase items listed in this roadmap:

Volcengine integration — Already landed. pkg/providers/factory_provider.go line 97 includes "volcengine" as a supported protocol, pkg/config/config.go has VolcEngine ProviderConfig, and getDefaultAPIBase() returns "https://ark.cn-beijing.volces.com/api/v3" (line 198). The config migration layer also handles legacy doubao model names (see pkg/config/migration.go:296). This item can be checked off.

Basic MCP support — Also landed. There is a working MCP client implementation in pkg/mcp/manager.go and pkg/tools/mcp_tool.go with tests. The manager handles stdio and sse transports, tool discovery, and call routing. The agent loop integrates MCP tools alongside built-in tools in pkg/agent/loop.go.

Multi-modal attachments — The MediaSender interface exists in pkg/channels/interfaces.go but is only implemented by a few channels. The channels manager already checks for MediaSender support at pkg/channels/manager.go:645 and skips media for channels that do not implement it. Expanding coverage to more channels (Feishu, DingTalk, QQ) would be the next step.

Agent Loop 2.0 — The current loop in pkg/agent/loop.go is a single-threaded iteration model. Parallel tool execution was added (multi-tool calls run concurrently), but the fundamental loop is still serial. Opportunities for latency reduction: streaming LLM responses (currently the Codex provider streams but the HTTP provider does not), and speculative tool prefetching.

Multi-Agent / Swarm — Subagent spawning exists (IsSubagentSessionKey() in pkg/routing/session_key.go), but there is no coordination protocol between agents. The session key structure already namespaces subagents, so the routing layer is ready for swarm coordination once the orchestration logic is built.

Looking forward to seeing progress on the mobile/Termux deployment. The Go binary cross-compiles cleanly to ARM64 — the main friction is the dependency on ffmpeg for voice transcription and any CGo-linked libraries.

[antoninadert]

Strong roadmap. A missing piece might be:

1. a tutorial or playground explaining the basic concepts behind picoclaw
2. A full list of up-to-date capabilities of pi-coclaw (what exactly it does and how it is architectured) -> a readme section
3. a few implementation guides with security best practices for docker and for normal installation (explaining also the potential bugs and failures that could happen to newbies)