gitaiflow / Architecture / Architecture Diagrams
DocsgitaiflowArchitectureArchitecture Diagrams

Architecture Diagrams

Detailed flow diagrams for individual gitaiflow subsystems. For the top-level pipeline, see gitaiflow.md.

4 min readApplies to v1.1.3
On this page ▾
  1. 1. Execution mode routing
  2. 2. Normal summary pipeline (detail)
  3. 3. Telemetry consent decision
  4. 4. Changelog generation
  5. 5. Diff caching / regeneration skip
  6. 6. Provider boundary

1. Execution mode routing

2. Normal summary pipeline (detail)

GITAIFLOW_TELEMETRY overriding the saved decision for "this process only" is what lets CI flip telemetry per-job without disturbing a developer's durable local answer — see ../user-guide/telemetry.md.

4. Changelog generation

Changelog generation never calls an AI provider and never records a normal-run usage or telemetry event — only Git and local artifact reads/writes are involved.

5. Diff caching / regeneration skip

text
DiffGenerationService generates diff for a group
        │
        ▼
diff content hash compared against
previously summarized diffs (any prior day)
        │
   ┌────┴────┐
   │         │
 match     no match
   │         │
   ▼         ▼
reuse       call AI provider,
earlier     write new summary
summary     for this diff
(no AI call)

--regenerate-summary bypasses this comparison unconditionally, forcing a fresh AI call for every group regardless of a matching prior diff.

6. Provider boundary

text
                AIClient
                   │
     OpenAI-compatible chat/completions
                   │
      ┌────────────┼─────────────────────────┐
      │            │                          │
   Gemini        OpenAI              custom endpoint
 (OpenAI-compat    /v1                (OpenRouter, Grok,
  endpoint)                            DeepSeek, vLLM,
      │                                LM Studio, ...)
      │                                        │
      ▼                                        ▼
  cloud provider                    cloud or self-hosted,
  receives prompt                   receives prompt + diff
  + diff content                    content
      │
      │            ┌───────────────┐
      └────────────│    Ollama     │◀── request stays on
                    │ (local model) │    the local machine
                    └───────────────┘

For local Ollama usage, the AI request never leaves the machine. For every cloud provider, the provider receives the filtered/redacted prompt and diff content required to generate the summary — see security-and-privacy.md for the full data-flow boundary.