FiboCode

Quick Start

Inside the editor, FiboCode builds a knowledge base from code, docs, and graphs—one that grows with the code—and shares that context with the Agent CLI you choose. This lets you and your agents understand, navigate, and change code together.

Understanding grows with the code.

Day to day that shows up as two ways of using it — one for writing, one for reading:

UsageWhat you're doingWhat FiboCode adds
Enhanced terminal workflowCoding through the Agent CLI you already trustWorkspace-aware context, automatically recorded file changes, IDE-grade hunk-by-hunk review
Analyzing and understanding codeTaking over an unfamiliar repository, or working out why a system became what it isWhole-repo explanation docs, concept graphs, and two-way jumps between code, text, and graph

Both paths start the same way: install the app, open a workspace, configure a model.

Download and Install

Grab the latest installer from the FiboCode Releases page:

OSPackage
macOS (Apple Silicon).dmg disk image
Windows.exe installer
Linux.AppImage
  • macOS: open the .dmg and drag FiboCode into Applications.
  • Windows: run the .exe installer and follow the prompts.

The Download button on the homepage always resolves to the newest release automatically, so you never have to hunt for version numbers.

Before You Start

1. Open a workspace. Click "Open Folder" in the workspace panel and pick your codebase, or drag a folder straight onto the panel. Dragging a .zip file is not supported; "Recently Opened" switches between workspaces later. FiboCode treats the repository, not a single file, as the unit of understanding.

2. Configure a model. Go to System Config → Model Config → Add Provider and fill in Provider Name, Interface Type (OpenAI compatible), Base URL, API Key, and Model Name; after saving, tick Main Model. Mark a low-latency provider as Fast Model so latency-sensitive calls prefer it. Advanced Settings covers concurrency, timeout, and max output tokens.

Once a model is in place, three capabilities come alive:

  • Diff summaries — hover a file or directory in the source control panel for an instant summary of what changed, and generate a commit message with one click before committing.
  • Code completion — the AI suggestion panel in the editor, accepted or rejected one suggestion at a time.
  • Code analysis — whole-repository analysis producing directory- and file-level explanation docs and concept graphs.

3. Install Fibo Skills for your agent (optional, recommended). Fibo Skills add a spec → plan → execute → sync workflow to the Agent CLI in your terminal and instruct it to write specifications, plans, and change records into the .fibo/ document system. FiboCode does not automatically create this evolution history; install and use Fibo Skills if you want the agent to maintain it. See Fibo Skills for installation.

Usage 1: The Enhanced Terminal Workflow

Keep the Agent CLI you already trust and change none of your habits — FiboCode only fills in the context and the review around it.

  1. Switch to CLI mode and start coding. Flip the chat panel into CLI mode, open the terminal, and run Claude Code, Codex, or whichever Agent CLI you prefer.
  2. Switch back to chat mode to settle the changes. When the session ends, switch back and FiboCode tallies every file touched in between, listing them on the changed files card — no more reconstructing from memory or git status what the agent actually did.
  3. Run a quick review. Hit Review on that card: read the diff hunk by hunk, accept or reject each block, or take a whole file at once. Snapshots let you roll back to an earlier state of a file.
  4. Commit. Open Source Control, stage the changes, generate or write the commit message, and commit.

Usage 2: Analyzing and Understanding Code

Reading a repository isn't about opening every file top to bottom — it's about letting the code describe its own structure.

  1. Configure scope and templates. Go to System Config → Code Analysis:
    • Directory Blacklist — directory names are matched exactly; a matched directory and everything under it is excluded (node_modules, dist, …).
    • Extension Whitelist — only files matching the code or configuration extension whitelist are analyzed.
    • Analysis Templates — the Directory, Code File, and Configuration File templates are edited separately and shape the structure of the generated explanations; each can be reset to its original content.
  2. Start the analysis. Back in the workspace panel, click Analyze to run the whole-repository pass. Directories and files are analyzed concurrently, bounded by the concurrency setting of your model config; you can abort at any time, and completion is announced when it finishes.
  3. Read the explanations. Click any directory or file and the Document view on the right shows its explanation; switch to the Graph view for the concept graph of that spot. Directory-level notes answer "what does this layer do", file-level notes answer "how does this file do it".
  4. Follow the tags. The code / text / graph tags inside an explanation are clickable anchors: code jumps to the source file and line range, text to the linked document, graph to the matching concept graph. Wherever you are reading, code and explanation stay reachable from each other.

Where to Go Next