Proxies API calls to Anthropic Claude.
API requests should include history
, a list of tuples of strings, and the current
query
Example Request:
{ "history": [["user", "hello"], ["llm", "hi"]], "query": "how are you?" }
Responses will be returned as dictionaries. Responses should contain the following:
response
- String LLM response to the query
When running this as a docker container, the XDG_CONFIG_HOME
envvar is set to /config
.
A configuration file at /config/neon/diana.yaml
is required and should look like:
MQ:
port: <MQ Port>
server: <MQ Hostname or IP>
users:
neon_llm_claude:
password: <neon_claude user's password>
user: neon_claude
LLM_CLAUDE:
key: ""
openai_key: ""
model: "claude-2"
role: "You are trying to give a short answer in less than 40 words."
context_depth: 3
max_tokens: 100
num_parallel_processes: 2
For example, if your configuration resides in ~/.config
:
export CONFIG_PATH="/home/${USER}/.config"
docker run -v ${CONFIG_PATH}:/config neon_llm_claude
Note: If connecting to a local MQ server, you may need to specify
--network host