AI-OPS Configuration¶
This is the full settings reference. For a guided setup see Getting Started, and for choosing a model see Model Selection.
API¶
The AI-OPS API is configured through the following environment variables:
| Variable | Type | Default | Required |
|---|---|---|---|
AI_OPS_HOST |
string | 127.0.0.1 |
No |
AI_OPS_PORT |
int | 8000 | No |
AI_OPS_AUTH_TOKEN |
string (secret) | - | No for localhost, Yes otherwise. |
AI_OPS_STORAGE_STRATEGY |
enum | JSONL |
No |
AI_OPS_MODEL |
string | - | Yes |
LLM_API_BASE |
string | - | No |
LLM_API_KEY |
string (secret) | - | No |
AI_OPS_MODEL, LLM_API_BASE, and LLM_API_KEY are covered in depth in
Model Selection. For launching the server and the auth
rules see Run the API Server, and for the full
HTTP surface see the API Reference.
CLI¶
The default CLI configuration is loaded from ~/.config/ai_ops/cli.json if the file exists, however the flags have precedence over the file.
Note: a missing file is not an error; a malformed file, or one that fails schema validation, is startup-fatal.
See Use the CLI for what each flag does in practice.
Flags¶
| Flag | Description |
|---|---|
--base-url <url> |
API base URL |
--api-key <key> |
API key, sent as X-AI-OPS-ApiKey on every request |
--mode <supervised\|unsupervised> |
Agent mode, set at startup only (no runtime toggle) |
--resume <short_id> |
Resume an existing conversation |
Configuration File¶
{
"base_url": "<url>",
"api_key": "<api_key>",
"mode": "[supervised|unsupervised]",
"resume": 1 or 0
}
Agent¶
Agent Settings¶
To configure the AI-OPS agent you can add ~/.local/share/ai_ops/agent_config.json:
| Key | Type | Default |
|---|---|---|
tools |
list[str] | DEFAULT_TOOLS |
context_view.kind |
raw | layered |
layered |
context_view.params |
dict | {max_window_tokens: LLM_MAX_CONTEXT_LENGTH or 8192} |
command_policies |
list[{kind, params}] | [] |
temperature |
float | 0.4 |
prompt_extension |
string | null | null |
confirmation_timeout_s |
float | 300.0 |
tools names come from Built-in Tools, or your own via
Add a Tool. context_view is described in
Context Management, and command_policies in
Command Policies. When running in a container, mount this
file as shown in
Run the API Server.
Core Settings¶
| Variable | Type | Default | Notes |
|---|---|---|---|
LLM_MAX_CONTEXT_LENGTH |
int | 8192 |
Also used as the default for agent_config.json's context_view.params.max_window_tokens |
AI_OPS_OBSERVABILITY_BACKEND |
string | - | Only mlflow supported |
MLFLOW_TRACKING_URI |
string | - | Required if observability backend is mlflow |
MLFLOW_TRACKING_USERNAME |
string | - | Required by the MLflow server; not enforced in _mlflow.py |
MLFLOW_TRACKING_PASSWORD |
string (secret) | - | Required by the MLflow server; not enforced in _mlflow.py |
MLFLOW_EXPERIMENT_NAME |
string | - | Optional |
MLFLOW_TRACKING_INSECURE_TLS |
bool | false |
Optional, not recommended |
AI_OPS_LOG_LEVEL |
string | - | Lowercase |
AI_OPS_LOG_FILE |
string (filename) | - | Created under ~/.local/share/ai_ops/ |
AI_OPS_LOG_STDOUT |
bool | - | "true" / "false" |
SKILL_VERIFY_INSTALLED |
bool | false |
If true, checks each skill's requirements against PATH at startup and exits on any miss |
The base directory that holds logs, sessions, the workspace, and skills is described in Storage Layout.