How the ai-memory Bootstrap Command Initializes a New Project with Existing Git History

The ai-memory bootstrap command ingests a project's Git commit history, documentation, and source files to generate an initial wiki manifest that seeds the AI-memory knowledge base.

The ai-memory bootstrap command in the akitaonrails/ai-memory repository transforms existing project artifacts into structured AI knowledge. When executed inside a Git repository, it automatically discovers the project root, collects historical commits and contextual files, and constructs a foundational bootstrap.md manifest through coordinated client-side collection and server-side LLM processing.

Step 1: Repository Discovery and Source Collection

The initialization process begins by establishing the project context and gathering raw materials from the local filesystem.

Auto-Discovering the Git Root

In crates/ai-memory-cli/src/commands/bootstrap.rs, the CLI first invokes discover_repo_root using libgit2 to locate the Git repository root (lines 45‑61). If detection succeeds, that directory becomes the source base; otherwise, the command falls back to the current working directory and automatically disables Git-commit collection.

Extracting Commits and Documentation

Once rooted, the CLI calls collect_sources (lines 81‑88) to assemble a vector of BootstrapSource objects. According to crates/ai-memory-consolidate/src/bootstrap.rs (lines 11‑13), this includes:

• Git commits (when include_git is true), represented as BootstrapSource values of kind GitCommit
• README files, documentation, and Rust module headers

The collect_sources logic resides primarily in crates/ai-memory-consolidate/src/collect_sources.rs, which walks the repository to extract these artifacts.

Step 2: Client-Side Token Management and Transmission

Before transmitting data to the server, the CLI optimizes the payload to prevent rejection.

Pruning to the Input Budget

To avoid HTTP 413 errors, prune_sources_to_budget (lines 90‑95) trims the source list to comply with the user-specified max_input_tokens limit. This client-side validation mirrors server-side constraints and ensures the request stays within LLM context windows.

Transmission to the Admin Endpoint

The CLI serializes the workspace, project name, and pruned source list into JSON and POSTs to POST /admin/bootstrap (lines 32‑44). This endpoint is defined in crates/ai-memory-mcp/src/admin.rs and handled by the server-side consolidator.

Step 3: Server-Side Knowledge Generation

Upon receiving the bundle, the server executes ai_memory_consolidate::bootstrap::process_sources to generate the knowledge base.

Duplicate Manifest Detection

The server first checks for an existing wiki/<workspace>/<project>/bootstrap.md at lines 298‑303 in crates/ai-memory-consolidate/src/bootstrap.rs. If the file exists and parses cleanly, the operation aborts unless the user provided the --force flag.

LLM Chunk Processing

The plan_bootstrap_chunks function (lines 351‑357) splits the collected sources into LLM-sized segments. The configured LLM (OpenAI or Anthropic) then generates concise summaries for each chunk, transforming raw Git history and documentation into structured content.

Creating the bootstrap.md Manifest

Finally, the system writes a bootstrap.md manifest via crates/ai-memory-wiki/src/wiki.rs, inserting YAML front-matter containing a bootstrapped_at timestamp and tags ["bootstrap","manifest"] (lines 430‑436). This manifest links the generated summary pages, establishing the project's initial knowledge graph. The server returns a BootstrapOutcome listing pages written and token usage, which the CLI renders as a human-readable report (lines 45‑50).

Implementation Reference: Key Source Files

• crates/ai-memory-cli/src/commands/bootstrap.rs – CLI entry point; discovers repo, collects sources, sends POST.
• crates/ai-memory-consolidate/src/bootstrap.rs – Server-side bootstrap logic; creates bootstrap.md, calls LLM.
• crates/ai-memory-consolidate/src/collect_sources.rs – Walks the repo, extracts Git commits and other sources.
• crates/ai-memory-mcp/src/admin.rs – HTTP handler for POST /admin/bootstrap that invokes the consolidator.
• crates/ai-memory-wiki/src/wiki.rs – Handles writing the generated wiki pages, including bootstrap.md.

Code Examples

// Example: Running bootstrap from a Git repository
let args = BootstrapArgs {
    repo_path: None,          // auto‑detect .git
    exclude_git: false,
    force: false,
    dry_run: false,
    ..Default::default()
};
ai_memory_cli::commands::bootstrap::run(&config, args).await?;

# Equivalent CLI invocation

AI_MEMORY_SERVER_URL=http://localhost:49374 \
AI_MEMORY_AUTH_TOKEN=... \
ai-memory bootstrap --chunk-input-tokens 2000 --max-input-tokens 150000

Summary

• The ai-memory bootstrap command initializes projects by transforming Git history into structured wiki pages.
• Repository discovery relies on libgit2 via discover_repo_root, falling back to the current directory if no .git exists.
• Source collection gathers Git commits, READMEs, and Rust module headers as typed BootstrapSource objects.
• Token pruning happens client-side to enforce max_input_tokens before transmission.
• Server-side processing generates LLM summaries and writes a bootstrap.md manifest with metadata tags.
• The process is idempotent by default; use --force to overwrite existing manifests.

Frequently Asked Questions

What happens if the project already has a bootstrap manifest?

The server checks for an existing wiki/<workspace>/<project>/bootstrap.md before processing. If the file exists and parses cleanly, the bootstrap operation aborts unless you supply the --force flag to overwrite the previous bootstrap state.

How does the command handle non-Git directories?

If discover_repo_root fails to find a Git repository, the CLI falls back to using the current working directory as the source base. In this mode, the tool disables Git-commit collection automatically, though it still processes other documentation files present in the directory.

What sources are included beyond Git history?

In addition to Git commits (when include_git is enabled), collect_sources ingests README files, general documentation, and Rust module headers. Each source type is tagged accordingly in the BootstrapSource enum to ensure proper handling during the LLM summarization phase.

How does the token budget affect the bootstrap process?

The prune_sources_to_budget function enforces the max_input_tokens limit client-side before transmission. This prevents HTTP 413 errors and ensures the payload fits within the LLM's context window, prioritizing recent or high-priority sources when trimming is required.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}