AI Backends

AI Backends

Parsing free text into fields is the only step in jasem that uses a model. Everything else — listing, reporting, the dashboard, editing — is pure local code. You pick a backend with environment variables, and if no model is reachable jasem still saves your entries using its built-in regex parsing (see Natural Language Input).

Three providers are supported, selected with JASEM_PROVIDER:

JASEM_PROVIDER	Backend	Needs a key?	Default model
ollama (default)	Local Ollama	no	qwen2.5:3b
openai	Any OpenAI-compatible API	yes	gpt-4o-mini
anthropic	Anthropic (Claude)	yes	claude-opus-4-8

Override the model for any provider with JASEM_MODEL.

Ollama — local, free, private (default)

The default backend. Nothing to configure beyond having Ollama running with a small model pulled:

ollama serve
ollama pull qwen2.5:3b

jasem talks to Ollama at http://localhost:11434 by default; point elsewhere with OLLAMA_HOST:

export OLLAMA_HOST=http://192.168.1.10:11434

Use any model you have locally:

export JASEM_MODEL=llama3.1:8b

This keeps everything on your machine — no API key, no network calls off-box.

OpenAI-compatible APIs

Works with OpenAI itself and any service that speaks the same API — Groq, OpenRouter, LM Studio, vLLM, and others:

export JASEM_PROVIDER=openai
export JASEM_API_KEY=sk-...
# For non-OpenAI hosts, set the base URL:
export JASEM_OPENAI_API_BASE=https://openrouter.ai/api/v1
export JASEM_MODEL=gpt-4o-mini          # or the host's model id

Base URL resolution — jasem reads the first of these that is set: JASEM_OPENAI_API_BASE → JASEM_OPENAI_BASE_URL → OPENAI_BASE_URL → JASEM_API_BASE. With none set it uses OpenAI's own endpoint.

API key resolution — JASEM_API_KEY → OPENAI_API_KEY → ANTHROPIC_API_KEY (the first that is set). So if you already export OPENAI_API_KEY, jasem picks it up automatically.

LM Studio / local OpenAI servers

export JASEM_PROVIDER=openai
export JASEM_OPENAI_API_BASE=http://localhost:1234/v1
export JASEM_API_KEY=lm-studio          # any non-empty string
export JASEM_MODEL=your-local-model

Anthropic (Claude)

export JASEM_PROVIDER=anthropic
export JASEM_API_KEY=sk-ant-...
export JASEM_MODEL=claude-opus-4-8       # optional; this is the default

(ANTHROPIC_API_KEY is also accepted as the key.)

What happens when the backend is down

jasem is designed to never lose an entry because a model is unavailable. If the backend is unreachable or misconfigured:

• Tasks are saved with the raw text as the title and the date parsed by regex.
• Time and spending fall back to a simple comma format: <duration/amount>, <description>[, <date>][, <tag>].

You'll see a short warning, but the add always succeeds. This means you can use jasem with no AI at all — you just type a touch more structure (commas) and write clean titles yourself.

# With the backend down, this still works:
jasem track "1h30min, code review, yesterday, work"
jasem acc   "50k, lunch, yesterday, food"

Checking your current backend

jasem help prints the resolved provider, model, and file paths at the bottom, so you can confirm what jasem will actually use:

provider          ollama   (JASEM_PROVIDER: ollama · openai · anthropic)
model             qwen2.5:3b   (JASEM_MODEL)

See also

• Configuration — the full environment-variable reference
• Natural Language Input — what the parser and offline fallback extract