Explore is an AI-assisted code exploration TUI. It pairs a hierarchical browser (repo → directory → file → symbol) with auto-generated AI explanations at every level, so the answer to "what is this?" is already on screen when you arrive. Navigation is vim-like / lazygit-like.

Implemented in Go / Bubble Tea.

Requirements

• Go 1.26+
• One of:
  • Claude (default): ANTHROPIC_API_KEY
  • OpenAI: OPENAI_API_KEY plus --provider openai
  • Ollama: a local Ollama server (defaults to http://localhost:11434), --provider ollama
• git on PATH — optional; enables the History tab and historical snapshots. Without it (or outside a git repo) the rest of the tool works unchanged.
• Language servers on PATH for caller/callee lookups (all optional; xref is disabled per-language when the server is missing):
  • Go: gopls
  • Python: pyright-langserver
  • TypeScript/TSX: typescript-language-server
  • Rust: rust-analyzer
  • Ruby: ruby-lsp
  • Java: jdtls
  • C/C++: clangd

Symbol-level explanations and tree-sitter syntax highlighting are available for Go, Python, TypeScript (.ts / .tsx), Rust, Ruby, Java, and C/C++. Other languages still show in the file tree but won't expand to a symbol list. The focused row drops styling so the cursor highlight reads cleanly.

Install

go install github.com/mikethicke/explore/cmd/explore@latest

Or from a clone:

go build -o explore ./cmd/explore

Usage

export ANTHROPIC_API_KEY=sk-ant-...
explore [path]                                # Claude (default)
explore --provider openai [path]              # OPENAI_API_KEY required
explore --provider ollama --model qwen2.5-coder:14b [path]

Flags:

• --cache-dir <dir> — override the BBolt cache location (default <repo>/.explore/cache.db)
• --provider <claude|openai|ollama> — pick an LLM backend (default claude)
• --model <id> — override the model (provider-specific default if empty)
• --ollama-host <url> — Ollama host (default: $OLLAMA_HOST or http://localhost:11434)
• --openai-endpoint <url> — OpenAI endpoint override (e.g. Azure-compatible proxy)
• --token-budget <N> — session token ceiling for the status-bar indicator; 0 means track-only
• --config <path> — use a specific config file instead of the default location
• --no-lsp — skip launching language servers
• --debug — write a debug log to <cache-dir>/debug.log

The status bar shows a running tok: 12.3k total once any LLM call lands. With --token-budget set, the figure becomes 12.3k/100k (12%) and turns yellow at 80%, red at 100%. Cached explanations (no new LLM call) don't add to the count. Before each request the payload is scanned for credential-shaped strings (gitleaks-style); a match shows a status-bar warning but never blocks or redacts the request.

Layout

Three panes:

• Tree (left) — repo → dir → file → symbol, expanded lazily. A second History tab on this pane lists commits.
• Explanation (center-top) — prose for the focused node, with Q&A (node) and Q&A (session) tabs alongside it.
• Source (center-bottom) — the focused file with syntax highlighting, line numbers, and line selection.

Explanations are generated bottom-up (symbols → files → dirs → repo) and content-hash cached, so re-opening anything is free until the source changes. A background prefetcher warms the nodes you're likely to visit next. The cache key includes the model name, so switching providers or models never returns stale explanations.

Q&A has two modes: node (a thread scoped to the focused node) and session (one running thread that auto-attaches whatever you're looking at, so you can ask follow-ups across nodes). Reach them via pane 2 or by cycling tabs with [ / ].

Git integration

Press H (or cycle to the History tab with [ / ] on the tree pane) to see the full commit history of the current branch.

• A synthetic WORKING row sits at the top, representing everything uncommitted (working tree vs. HEAD, including untracked files).
• Highlighting a commit shows its message and changed-file list in the source pane, plus an AI explanation of what that commit changed in the explanation pane.
• Enter on a commit drops the whole app into that commit's state: the tree, source, and explanations all reflect the repo as it existed then (read from git objects — your working tree is never touched). Changed nodes are colored (green = added, yellow = modified) and the source pane shows the full file with the commit's additions/removals highlighted inline. Enter on WORKING does the same against your uncommitted changes.
• Esc or b returns to the live working tree. b / Ctrl+O / Ctrl+I navigate across the live↔historical boundary correctly.

LSP cross-references are disabled inside a historical snapshot (language servers index the working tree, not history); they stay live in WORKING mode.

Config file

Settings can be persisted to ~/.config/explore/config.toml (or $XDG_CONFIG_HOME/explore/config.toml). CLI flags override file values; file values override defaults.

[provider]
default = "claude"

[provider.claude]
model = "claude-sonnet-4-6"
api_key_env = "ANTHROPIC_API_KEY"

[provider.openai]
model = "gpt-4o-mini"
api_key_env = "OPENAI_API_KEY"
# endpoint = "https://my-azure-proxy.example.com/v1/chat/completions"

[provider.ollama]
model = "qwen2.5-coder:14b"
host = "http://localhost:11434"

[ui]
token_budget = 0              # 0 = track only, no ceiling
no_lsp = false
long_function_threshold = 200 # functions longer than this get a structured outline; 0 disables

Use --config <path> to point at a different file. A missing file is fine (defaults apply); a malformed file is fatal so typos surface immediately.

Sharing the cache

Explanations can be exported and imported so a team shares LLM output instead of regenerating it. Both are one-shot subcommands that exit before the TUI starts:

explore export-cache cache.json [repo-path]
explore import-cache [--overwrite] cache.json [repo-path]

Import skips entries that already exist locally unless --overwrite is given. Because cache keys encode the source hash, level, prompt version, and model, imported entries only ever apply to byte-identical source under the same model.

Keys

Navigation is vim-ish. Press ? for the full in-app cheat sheet.

Panes & tabs

• 1 / 2 / 3 — focus tree / explanation / source
• Tab / Shift+Tab — next / previous pane
• [ / ] — previous / next tab within the focused pane
• o / b / Ctrl+O — back in the nav stack; Ctrl+I — forward

Tree pane

• j / k / arrows — move; Space / l / → expand, h / ← collapse or climb to parent
• gg / G — top / bottom; Ngg — jump to row N (prefix any motion with a count, e.g. 10j)
• Enter — open the node in the source pane
• H — switch to the History tab (commits + WORKING)

Source pane

• j / k — line down / up; J / K or Ctrl+D / Ctrl+U — page
• gg / G — first / last line; Ngg or :N — jump to line N
• v — start/stop line selection (extend with j/k); x — explain the selection (or current line)

Actions

• / — fuzzy search files & symbols; ↑/↓ to pick, Enter to jump, Esc to close
• r — regenerate the current explanation (bypasses the in-memory and on-disk caches)
• e — open the focused file in $EDITOR
• u — callers of the focused symbol; d — callees on the current source line (picker when >1)
• y — yank menu: p path · e explanation · s source
• ? — full keyboard cheat sheet; q or Ctrl+C — quit

License

MIT — see LICENSE.