Use Infersia with aichat

Setting it up

aichat keeps everything in one config file, and adding a provider means editing it. That is true whether or not you have used aichat before — the setup wizard only runs when there is no config file at all.

aichat --info | grep config_file

model: infersia:deepseek/deepseek-v4-flash-0731

clients:
  - type: openai-compatible
    name: infersia
    api_base: https://api.infersia.com/v1
    api_key: isk-v1-your-key-here
    models:
      - name: deepseek/deepseek-v4-flash-0731
        max_input_tokens: 1048576
        supports_function_calling: true
      - name: qwen/qwen3.6-35b-a3b
        max_input_tokens: 262144
        supports_function_calling: true
        supports_vision: true
      - name: qwen/qwen3-14b
        max_input_tokens: 131072
        supports_function_calling: true
      - name: qwen/qwen3-8b
        max_input_tokens: 40960
        supports_function_calling: true
      - name: qwen/qwen3-8b:free
        max_input_tokens: 40960
        supports_function_calling: true

model: infersia:qwen/qwen3-8b:free

clients:
  - type: openai-compatible
    name: infersia
    api_base: https://api.infersia.com/v1
    api_key: isk-v1-your-key-here
    models:
      - name: qwen/qwen3-8b:free
        max_input_tokens: 40960

aichat --list-models          # should print five infersia: entries
aichat "Say hello in exactly three words."

Installing aichat for the first time?

If you have no config file yet, running aichat offers to build one. Answer yes, then pick openai-compatible from the provider list — it is the second entry, above all the named platforms. It then asks for four things:

Provider Name:  infersia
API Base:       https://api.infersia.com/v1
API Key:        isk-v1-your-key-here
LLMs to include: (multi-select, fetched from your endpoint)

That last step calls /v1/models and offers our catalogue as a checklist, so you don’t have to type model ids. What it cannot fill in is max_input_tokens — so once the wizard has written the file, open it and add the limits from the block above.

The wizard is genuinely first-run only. There is no command that adds a second provider later, so if aichat is already configured for something else, ignore this section and edit the file.

Switching models

Every model is addressed as infersia: plus its id, including the publisher prefix:

# one-off, from your shell
aichat -m infersia:qwen/qwen3-8b:free "summarise this"

# inside the REPL
.model infersia:qwen/qwen3.6-35b-a3b

The model: key at the top of the config sets the default when you don’t pass -m.

The shell assistant

This is what most people install aichat for, and it is the workload our pricing suits best — hundreds of very short requests a day, where a cheap model is indistinguishable from an expensive one:

aichat -e "find every file over 100MB under my home directory"
aichat -c "a bash function that retries a command three times"

qwen/qwen3-8b is a good default here — a one-line shell command does not need a 284B model, and at $0.05 per million input tokens a heavy day of -e costs less than a cent. Set it per-invocation with -m, or keep it as the default and reach for DeepSeek when you’re piping in real files.

Which model to pick

If something goes wrong

“Incorrect API key provided”

Check the key was pasted whole — they are long, and a partial copy is the usual cause. Confirm it independently of aichat:

curl https://api.infersia.com/v1/models \
  -H "Authorization: Bearer isk-v1-your-key-here"

A JSON list of models means the key is good and the problem is in the config.

--list-models shows nothing from us

Almost always YAML indentation. The models: list belongs inside the client entry, at the same level as api_base — not at the top of the file. Compare against the block above, and check aichat --info is reading the file you edited.

“Model not found”

Model ids include the publisher prefix, and the reference includes the client name on top of that — infersia:qwen/qwen3-8b, not qwen3-8b. The exact strings are on the catalogue page, each with a copy button.

Sessions compress sooner than you expect

That is max_input_tokens missing or set low — see the note above. aichat also has its own compress_threshold, which defaults to 4,000 tokens and is independent of the model’s window; raise it in the config if you want long sessions kept verbatim.

Does --sync-models overwrite this?

No. That command refreshes aichat’s own registry of built-in providers, which is separate from the client you defined. We checked: running it leaves this configuration untouched.

aichat is an open-source project by sigoden, published under the Apache 2.0 and MIT licences. Its name is used here only to identify the software these instructions are for. Infersia is not affiliated with or endorsed by the project.