DailyRecap
Turn a day of AI-assisted coding into a decision retrospective — what you decided, why, and what it produced — with one command.
Retrospectives are valuable but rarely happen, because writing them is friction. DailyRecap reads your AI coding session logs (and git history) and drafts the recap for you, so the day's decisions and context accumulate with almost no effort. It's a private journal for you, not a publishing tool.
Features
- Decision-first recaps. Each day is organized into work threads (e.g. Sign-up, UI fixes, Backend API), and every thread gets a structured section: what was decided → why (context & judgment) → result → what got stuck → what's next.
- Recovers the "why" from your session. It reads Claude Code session logs (your prompts, the assistant's replies, the tool calls) to reconstruct the reasoning behind each change — not just the diff.
- Fuses in git results. The same day's commits and changed files are pulled in as ground-truth evidence for the "result" of each decision.
- Grounded, not hallucinated. Facts and numbers must trace to a source; only choices you explicitly made are stated as decisions; quotes carry citations.
- Writes where you already work. Appends to local Markdown or an Obsidian vault, organized by date.
- Your choice of model, your key (BYOK). Use Anthropic, OpenAI, Gemini, Grok, or a local Ollama model. Cloud keys live only in the editor's secret storage and calls go directly to the provider — never through our servers. Ollama runs entirely on your machine.
Requirements
- VS Code 1.85+ (or Cursor / other VS Code-compatible editors)
- A model to generate with — either:
- an API key for a supported cloud provider (Anthropic, OpenAI, Google Gemini, or xAI Grok), with usage credit billed by that provider, or
- a local Ollama install (free, offline, no key)
- Node.js 18+ is only needed if you build the extension from source
Getting Started
- Pick a provider (see Providers). The default is Anthropic.
- Cloud provider: open the Command Palette (
Cmd/Ctrl+Shift+P) → DailyRecap: Set API Key, choose the provider, and paste its key (stored in SecretStorage).
Ollama (local): no key — just set recap.provider to ollama and recap.model to a local tag (see Using Ollama).
- Run DailyRecap: Generate Recap.
- Pick a project → a date (dates that actually have logs are listed first) → confirm.
- Review the preview, then choose where to save it (Local Markdown or Obsidian).
No Claude Code logs for that day? DailyRecap offers to build the recap from that day's git commits instead.
Providers
DailyRecap is BYOK and provider-agnostic. Choose one via the recap.provider setting and set its model in recap.model.
| Provider |
recap.provider |
API key |
Default base URL |
Example recap.model |
| Anthropic (Claude) |
anthropic |
required |
https://api.anthropic.com |
claude-sonnet-4-6 |
| OpenAI (GPT) |
openai |
required |
https://api.openai.com/v1 |
gpt-4o |
| Google Gemini |
gemini |
required |
https://generativelanguage.googleapis.com/v1beta/openai |
gemini-2.5-pro |
| xAI (Grok) |
grok |
required |
https://api.x.ai/v1 |
grok-4 |
| Ollama (local) |
ollama |
none |
http://localhost:11434/v1 |
llama3.1 |
Model names above are examples — check each provider's docs for current names.
Grounding note. Anthropic uses the native Citations API to tie extracted facts and quotes back to your source material. OpenAI, Gemini, Grok, and Ollama have no Citations API, so they run through an inline text-extraction path instead — recaps still work, but their grounding is slightly weaker (local models most of all).
Managing keys
Each cloud provider has its own key slot, so you can store several and switch between them by changing recap.provider — DailyRecap never has to guess which key is which.
- DailyRecap: Set API Key — pick a provider and store/replace its key.
- DailyRecap: Manage API Keys — see which providers have a key stored (● stored / ○ none), set/replace by selecting a row, or delete with the trash icon. (Ollama isn't listed because it has no key.)
Keys are stored only in the editor's secret storage (e.g. macOS Keychain) — never in settings.json, logs, or any server.
Using Ollama (local)
Ollama needs no API key; it's configured entirely through settings.
Install Ollama from ollama.com and start it (ollama serve).
Pull a model, e.g. ollama pull llama3.1 (or qwen2.5-coder, etc.).
In VS Code settings (settings.json or the Settings UI):
{
"recap.provider": "ollama",
"recap.model": "llama3.1", // must match a model you've pulled
"recap.baseUrl": "" // empty → http://localhost:11434/v1
}
⚠️ When switching to Ollama, remember to change recap.model — the default (claude-sonnet-4-6) is an Anthropic model and won't exist locally. If the local server isn't running, DailyRecap will tell you to start it.
Commands
| Command |
ID |
Description |
| DailyRecap: Set API Key |
recap.setApiKey |
Pick a provider and store its API key in SecretStorage |
| DailyRecap: Manage API Keys |
recap.manageApiKeys |
List which providers have a stored key; set/replace, or delete via the trash icon |
| DailyRecap: Generate Recap |
recap.generate |
Run the full flow: pick project/date → generate → preview → record |
Extension Settings
| Setting |
Default |
Description |
recap.provider |
anthropic |
LLM provider: anthropic, openai, gemini, grok, or ollama. See Providers. |
recap.model |
claude-sonnet-4-6 |
Model for the active provider. Change this when you switch providers. |
recap.baseUrl |
(unset) |
Override the base URL for OpenAI-compatible providers (openai/gemini/grok/ollama). Empty uses the provider default. |
recap.source |
session |
Raw-material source: session (Claude Code logs) or git. |
recap.sink |
local |
Where to record: local or obsidian. |
recap.outputDir |
./recaps |
Output directory for local Markdown. |
recap.obsidianVault |
(unset) |
Absolute path to your Obsidian vault. |
recap.lang |
auto |
Recap output language (auto/ko/en); auto follows the editor locale. |
recap.gitEnrich |
always |
Enrich the "result" axis with the same day's git commits: always / auto (only when the session's own result signal is weak) / off. |
Recording paths — Local: <outputDir>/YYYY/MM/DD-recap.md, Obsidian: <vault>/_Recap/YYYY/MM/DD.md. Re-running the same day appends with a separator; identical content is skipped.
How it works
DailyRecap makes a two-stage LLM call: (1) it extracts decisions, quotes, and numbers from the tagged source material, then (2) generates the topic-segmented recap from that. With Anthropic, stage 1 uses the Citations API for stronger grounding; the other providers use an inline text-extraction path instead. The two stages are separate because Citations and Structured Outputs can't be combined in one call.
Source material is tagged by origin — your prompts, the assistant's replies, tool calls/results, and (when enriched) git commits and changed files — and trimmed to fit a token budget, prioritizing your prompts and the assistant's reasoning over raw tool output.
Privacy & Security
Your API key is stored only in the editor's secret storage (e.g. macOS Keychain) — never in settings.json, logs, or any server. Before raw material is sent to the model, keys, tokens, and emails are masked. Cloud calls go directly to the provider you configured; with Ollama, nothing leaves your machine. The preview webview runs with a strict CSP and no scripts.
Known limitations
- Auto-matching a workspace to its
~/.claude/projects folder can fail when the folder-name encoding differs across Claude Code versions — pick the project manually from the list when that happens.
- The "why" reconstructed from git-only days is only as good as your commit messages.
- Anthropic is the only provider with the native Citations API; OpenAI/Gemini/Grok and local Ollama run through an inline extraction path instead, so their grounding is slightly weaker (local models most of all).
Contributing
Contributions welcome — see CONTRIBUTING.md.
Release Notes
See CHANGELOG.md.
License
MIT
