Multi-model AI assistant

how to run

# interactive chat
uv run ai.py chat

# chat with anthropic and a named session
uv run ai.py chat -p anthropic -s my-session

# analyse a file
uv run ai.py analyse report.pdf

# analyse with more detail and use anthropic
uv run ai.py analyse notes.txt -p anthropic --detail detailed

# compare both providers on the same prompt
uv run ai.py compare "explain how neural networks work"

# see your cost breakdown
uv run ai.py costs

---

this project is part of my AI learning journey.

in this project I built a CLI tool to chat interactively with OpenAI and Anthropic. I used the builder pattern with an abstract base class that all AI providers need to implement, this way I can switch between models easily without affecting the business code.

I used Typer to read CLI arguments and commands.

we have 4 commands

1. chat — start an interactive chat session.
2. analyse — analyse a .txt or .pdf file and print a structured summary.
3. compare — send the same prompt to both providers and show responses side by side.
4. costs — show cost breakdown from logged API calls.

flags

each command has its own flags but there are some shared ones:

--provider / -p — pick openai or anthropic, defaults to openai
--model / -m — override the default model if you want a specific one
--session / -s — name for the session, defaults to default, session history is saved to a json file so you can pick up where you left off
--system — custom system prompt, defaults to "You are a helpful assistant."
--stream — stream the response token by token instead of waiting for the full reply

the analyse command has a couple extra ones:

--detail — how thorough the summary should be, options are brief, standard, or detailed, defaults to standard
--json — dump the raw json output instead of the pretty printed version

---

setup

you need python and uv installed

git clone <repo>
cd multi-model-ai-assistant
uv sync

then create a .env file in the root with your api keys:

OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

---

how it works

the core idea is a simple abstract base class in llm/base.py that defines what any AI provider needs to implement — basically just a complete() method. then openai_client.py and anthropic_client.py each implement that.

this means the command code in ai.py doesn't care which provider you picked, it just calls client.complete() and gets back an LLMResponse object with the text, token counts, and cost. adding a new provider later would just mean writing one new file.

every request gets logged to cost_log.jsonl so the costs command can show you a breakdown of what you've spent across providers and commands.

the compare command runs both providers in parallel using asyncio so you're not waiting for them one after the other.

---

project structure

ai.py              # cli entry point, all 4 commands live here
llm/
  base.py          # abstract base class for providers
  openai_client.py
  anthropic_client.py
tools/
  costs.py         # logging and cost dashboard
  document.py      # pdf/txt reading and summary schema
sessions/          # saved chat histories (json files)
cost_log.jsonl     # append-only log of every api call