Setting up your LLM

Which should I pick?

Not sure? Start with Ollama if you value privacy and zero cost, or Anthropic if you want the best writing and don't mind paying cents per session.

---

Option A — Ollama (local & free)

Ollama runs an open language model entirely on your own computer. No account, no key, no bill, and it works offline once the model is downloaded. It's the most private option.

1. Install Ollama

   Download it for your OS from ollama.com/download and run the installer. On macOS and Windows it runs in the background automatically.

2. Pull a model

   Open a terminal and download a model. A small, capable starting point:

   ollama pull gemma4:e2b

   Models we'd recommend for session summaries, roughly smallest to best:

   Model	RAM	Good for
   gemma4:e2b	~8 GB	Lightweight machines; quick, decent summaries
   gemma4:e4b	~16 GB	The sweet spot — strong multilingual summaries (incl. German), still local
   qwen3:30b (A3B Instruct)	~16 GB+	Faster and more detailed than gemma4 thanks to its mixture-of-experts design

   Browse more in the model library. For summarizing, prefer an Instruct model over a Thinking / reasoning one — thinking models are slower and can leak their reasoning into the summary.

3. Make sure it's running

   Ollama serves an API at http://localhost:11434. If the app can't reach it, start it manually:

   ollama serve

4. Point Chronicle Keeper at it

   In Settings → LLM, choose Ollama, confirm the base URL is http://localhost:11434, and pick the model you pulled. Hit Test to confirm the connection.

Long sessions & speed

A full 3–4 hour session is a lot of text. Chronicle Keeper automatically sizes the model's context window to fit your whole transcript, so the summary always reflects the entire session — not just the first part. You don't need to configure anything.

Two advanced settings let you tune the local behaviour if needed (most people never touch them):

• Context window cap (ollama_num_ctx_max) — the largest window the app will request. The default fits a long session; lower it only if a big model spills onto your CPU and crawls.
• Timeout (ollama_timeout_seconds) — how long to wait before giving up. The default (several minutes) covers long local runs; raise it on slower machines or for very long sessions.

---

Option B — a cloud API key

Cloud models often write the best recaps. You'll create an account with a provider, generate an API key (a long secret string), and paste it into Chronicle Keeper's Settings. The key is stored locally on your machine.

Anthropic (Claude)

Built-in via the native Messages API. Great default for quality.

1. Create an account

   Sign up at console.anthropic.com and add billing credit.

2. Generate a key

   Go to Settings → API Keys → Create Key, name it (e.g. "Chronicle Keeper"), and copy the sk-ant-… string. It's shown only once.

3. Paste it in

   In Chronicle Keeper Settings → LLM, choose Anthropic, paste the key, pick a model (e.g. a current Claude), and hit Test.

OpenAI-compatible providers

One built-in client covers OpenAI and a family of compatible services. Pick whichever you already have an account with — the setup is the same: create a key, paste it, select the model.

General steps for any of them

1. Sign up & add billing

   Create an account on the provider's site (links above) and add a payment method or claim the free tier.

2. Create an API key

   Find the API keys section, create a new key, and copy it. Keys are usually shown only once — store it somewhere safe.

3. Add it in Settings

   In Chronicle Keeper, open Settings → LLM providers, pick the provider, paste the key, choose a model, and click Test to verify it works.

Test before you transcribe

Whichever route you choose, use the Test / Ping button next to the provider in Settings. A green result means the key and model are good and you're ready to summarize. If it fails, see Troubleshooting → LLM.