v1.0.0

dipper-bot v1.0.0

dipper-bot a compact, self-hosted personal AI assistant built as a single Go service with integrated tools, memory, channels, and web chat.

Highlights

Core Agent Platform

• Single-process architecture for gateway + agent loop + tools + optional web UI.
• Session-safe execution: new messages cancel in-flight runs for the same session.
• Optional per-run timeout controls for operational safety.

Rich Tooling Out of the Box

• File operations: read/write/edit/list with optional workspace restriction.
• Shell execution with timeout controls.
• Web search and web fetch.
• Messaging, session list/history/send, cron scheduling, and spawn/delegation tools.
• MCP integration (tools.mcpServers) for external tool ecosystems.
• Browser automation via Chrome DevTools MCP.

Long-Context & Memory System

• LCM (Lossless Context Management) with DAG + SQLite context assembly.
• LCM enabled by default (agents.defaults.lcm.enabled=false to disable).
• Token-based memory consolidation (contextWindowTokens) into:
  • memory/MEMORY.md
  • memory/HISTORY.md
• Curated memory via memory tool for:
  • memory/USER.md
  • memory/NOTE.md
• Session search over FTS index (memory/sessions_fts.db) via session_search.

Autonomous Learning Loops

• Background memory maintenance loop (memoryMaintenance) with constrained actions and quality/confidence gates.
• Background skills evolution loop (skillsEvolution) to create/patch skills/*/SKILL.md.
• Optional mid-run skill reflection for long tool loops:
  • midRunReflectEveryToolIters
  • midRunReflectMinSeconds

Learner Feedback UX

• User-visible learning feedback for autonomous memory/skill writes.
• Two mutually exclusive feedback modes:
  • Immediate push (default): experience.learnerFeedbackInstantPush=true
  • Next-reply digest: set experience.learnerFeedbackInstantPush=false

Channels & Interfaces

• Built-in channels: Telegram, WhatsApp (Baileys bridge), Discord, Feishu, DingTalk, Slack, Email, QQ, Wecom, Webhook.
• HTTP gateway (POST /message) for integrations.
• Built-in web chat UI (agent --web, default http://localhost:8600).
• Learning dashboard API in web UI (GET /api/learning/dashboard).

Security & Reliability

• Request size limits and strict JSON decoding on ingress.
• Safer web fetch defaults (private/loopback protections).
• Configurable workspace sandboxing (tools.restrictToWorkspace, default true).
• Rotating structured logs with process-wide logging controls.

Configuration & Operations

• Clear default/explicit semantics using pointer-backed config toggles (omitted vs explicit true/false).
• Expanded operational controls for learning loops, logging, usage accounting, and model pricing overrides.
• Workspace-first design with clean separation of:
  • user config
  • session storage
  • memory files
  • cron data
  • uploads

Documentation

The release includes updated documentation across:

• README.md / README_CN.md
• doc/MANUAL.md / doc/MANUAL_CN.md
• doc/CONFIG.md / doc/CONFIG_CN.md
• doc/DESIGN.md / doc/DESIGN_CN.md
• doc/LCM.md / doc/LCM_CN.md
• doc/PRIVACY.md / doc/PRIVACY_CN.md

Notes

• This release is designed for self-hosted, workspace-first usage with auditable behavior and low operational footprint.
• For detailed field-by-field configuration, use doc/CONFIG_CN.md (full reference) and doc/CONFIG.md (English concise reference).