Skip to content

v2.6.7-beta

Pre-release
Pre-release

Choose a tag to compare

View all tags
@codewithshinde codewithshinde released this 29 Jun 05:40
· 195 commits to main since this release
v2.6.7-beta
e4595a5

Full Changelog: https://github.com/Mitii-dev/Mitii/commits/v2.6.7-beta

Features

Context and indexing

  • Workspace scanner respects .gitignore and .thunderignore; auto-indexes on folder open
  • FTS5 full-text search with ripgrep fallback for unindexed paths
  • Symbol extraction for TypeScript, JavaScript, Python, Java, Go
  • PageRank repo map — surfaces the files that matter structurally, not just lexically
  • Vector search — local MiniLM embeddings via @xenova/transformers, hash fallback if unavailable; SQLite or LanceDB backend
  • Hybrid retriever + reranker — merges search, vectors, mentioned files, git state, and LSP diagnostics into one context pack

Agent workflow

  • Plan / Act / Review modes — plan before you touch code; act with tools; review without rewriting
  • Tool loop — read_file, search, write_file, apply_patch, run_command, git_diff, diagnostics, and more
  • Research subagents — parallel read-only workers for exploration (spawn_research_agent)
  • Task decomposition — multi-step plans with lifecycle tracking and step completion markers
  • Post-edit verification — configurable lint/test commands after Act-mode runs
  • Skills catalog — drop SKILL.md files in .thunder/skills/ and invoke them via use_skill

Safety and control

  • Approval modes — review_all, ask_edits, ask_deletes, ask_commands, auto
  • Autonomy presets — safe, guided, builder, pilot, enterprise (network disabled)
  • Untrusted workspace blocking — writes and shell disabled unless you explicitly opt in
  • Auto-checkpoints before approved file writes
  • Optional diff preview in VS Code before patches land
  • Patch validation — syntax guards before apply; refuses shell commands masquerading as source code

Memory and persistence

  • Long-term memory — memory_search / memory_write with FTS5 + optional vector hybrid search
  • Post-task memory extraction — observations captured after completed work
  • Session history — agent_sessions and agent_turns stored in SQLite; resume from the History tab
  • Plan persistence — plans saved to task_plans and .thunder/tasks/

MCP and integrations

Preloaded keyless MCP servers (disable with thunder.mcp.preloadBuiltin: false):

Server | Purpose -- | -- filesystem | Scoped file access for the open workspace memory | Cross-session knowledge graph sequential-thinking | Structured reasoning helper

Add your own via thunder.mcp.servers, .thunder/mcp.json, or .mcp.json. MCP tools appear as mcp__server__tool and still pass through Mitii's approval policy.

UI

React sidebar webview with chat, history, settings, approval cards, subagent activity panel, plan panel, checkpoint browser, indexing status, context budget warnings, and token meter.

Assets 2
Loading