notilens 0.2.0

NotiLens — unified SDK + CLI for AI agent notifications

NotiLens

Unified SDK + CLI for sending notifications from AI agents and any Python project to NotiLens.

Installation

pip install notilens

With framework auto-patching:

pip install notilens[openai]
pip install notilens[anthropic]
pip install notilens[all]        # all frameworks

---

Quick Start

CLI (hooks, shell scripts)

# Register your agent
notilens init --agent my-agent --token YOUR_TOKEN --secret YOUR_SECRET

# Use in scripts / Claude Code hooks
notilens task.start --agent my-agent --task job_001
notilens task.complete "All done" --agent my-agent --task job_001

Python SDK

import notilens

agent = notilens.init(agent="my-agent", token="YOUR_TOKEN", secret="YOUR_SECRET")

agent.task_start(task_id="job_001")
agent.task_complete("All done", task_id="job_001")

Python SDK with auto-patching

import notilens

agent = notilens.init(
    agent="my-agent",
    token="YOUR_TOKEN",
    secret="YOUR_SECRET",
    patch=True,   # auto-instruments OpenAI, Anthropic, LangChain, CrewAI, etc.
)

# Use OpenAI, Anthropic etc. normally — events fire automatically

---

Setup

Get your token and secret from the NotiLens dashboard.

CLI

notilens init --agent AGENT --token TOKEN --secret SECRET

SDK (env vars)

export NOTILENS_TOKEN=your_token
export NOTILENS_SECRET=your_secret

agent = notilens.init(agent="my-agent")  # reads from env

Multiple agents

notilens init --agent scraper --token TOKEN_A --secret SECRET_A
notilens init --agent mailer  --token TOKEN_B --secret SECRET_B

scraper = notilens.init(agent="scraper", token="TOKEN_A", secret="SECRET_A")
mailer  = notilens.init(agent="mailer",  token="TOKEN_B", secret="SECRET_B", patch=True)

Only one agent can own patch=True — raises if more than one tries.

---

CLI Reference

Task Lifecycle

notilens task.start    --agent AGENT [--task ID]
notilens task.progress "msg"  --agent AGENT --task ID
notilens task.loop     "msg"  --agent AGENT --task ID
notilens task.retry    --agent AGENT --task ID
notilens task.stop     --agent AGENT --task ID
notilens task.complete "msg"  --agent AGENT --task ID
notilens task.error    "msg"  --agent AGENT --task ID
notilens task.fail     "msg"  --agent AGENT --task ID
notilens task.timeout  "msg"  --agent AGENT --task ID
notilens task.cancel   "msg"  --agent AGENT --task ID
notilens task.terminate "msg" --agent AGENT --task ID

Input / Human-in-the-loop

notilens input.required "Please confirm" --agent AGENT --task ID
notilens input.approve  "Approved"       --agent AGENT --task ID
notilens input.reject   "Rejected"       --agent AGENT --task ID

AI Response

notilens ai.response.generate "Summary ready" --agent AGENT --task ID
notilens ai.response.fail     "Model error"   --agent AGENT --task ID

Generic Event

Emit any custom event — works for non-AI projects too:

notilens emit user.registered "New signup" --agent AGENT
notilens emit disk.space.full "Only 2GB left" --agent AGENT
notilens emit order.placed "Order #1234" --agent AGENT

Agent Management

notilens agents                           # list configured agents
notilens remove-agent AGENT              # remove an agent
notilens version

---

SDK Reference

notilens.init()

agent = notilens.init(
    agent="my-agent",         # agent name (required)
    token="...",              # NotiLens token (or NOTILENS_TOKEN env var)
    secret="...",             # NotiLens secret (or NOTILENS_SECRET env var)
    patch=False,              # auto-patch AI frameworks
    min_level="info",         # minimum event level to send
    loop_threshold=10,        # AI calls within loop_window before loop alert
    loop_window=60.0,         # time window in seconds for loop detection
    call_timeout=30.0,        # alert if AI call exceeds this many seconds
    silent=False,             # suppress SDK log output
    debug=False,              # enable verbose logging
)

Task Lifecycle

task_id = agent.task_start(task_id="job_001")   # task_id auto-generated if omitted
agent.task_progress("Step 2 of 5", task_id=task_id)
agent.task_loop("Processing item 42", task_id=task_id)
agent.task_retry(task_id=task_id)
agent.task_stop(task_id=task_id)
agent.task_complete("Done", task_id=task_id)
agent.task_error("Step failed", task_id=task_id)
agent.task_fail("Unrecoverable", task_id=task_id)
agent.task_timeout("Exceeded limit", task_id=task_id)
agent.task_cancel("User cancelled", task_id=task_id)
agent.task_terminate("OOM", task_id=task_id)

Input / Human-in-the-loop

agent.input_required("Confirm before proceeding", task_id=task_id)
agent.input_approved("User confirmed", task_id=task_id)
agent.input_rejected("User rejected", task_id=task_id)

AI Response

agent.ai_response_generated("Summary ready", task_id=task_id)
agent.ai_response_failed("Model unavailable", task_id=task_id)

Generic Emit

agent.emit("user.registered", "New signup", meta={"plan": "pro"})
agent.emit("disk.space.full", "Only 2GB left", level="warning")
agent.emit("order.placed", "Order #1234", meta={"amount": 99.99})

---

CLI Options

Flag	Description
--agent NAME	Agent name (required)
--task ID	Task ID (auto-generated if omitted)
--level	Override level: debug info warning error
--meta key=value	Custom metadata (repeatable)
--image_url URL	Attach an image
--open_url URL	Link to open
--download_url URL	Link to download
--tags "tag1,tag2"	Comma-separated tags
--is_actionable true|false	Override actionable flag
--confidence 0.0-1.0	Confidence score

---

Events

Event	Level	Description
task.started	info	Task began
task.progress	info	Mid-run update
task.loop	warning	Loop iteration
task.retrying	warning	Retry attempt
task.completed	info	Task finished successfully
task.stopped	info	Manually stopped
task.failed	error	Task failed
task.error	error	Non-fatal error
task.timeout	error	Exceeded time limit
task.cancelled	warning	Task cancelled
task.terminated	error	Force-terminated
ai.call.start	info	AI framework call began
ai.call.complete	info	AI framework call finished
ai.response.generated	info	AI produced a response
ai.response.failed	error	AI failed to respond
input.required	warning	Waiting for human input
input.approved	info	Input approved
input.rejected	warning	Input rejected

---

Full Example

# Register
notilens init --agent summarizer --token my_token --secret my_secret

# Run a job
notilens task.start --agent summarizer --task job_42
notilens task.progress "Fetching document" --agent summarizer --task job_42
notilens task.complete "Summary ready" --agent summarizer --task job_42 \
  --open_url https://example.com/summary.pdf \
  --meta pages=12

import notilens
import openai

agent = notilens.init(
    agent="summarizer",
    token="my_token",
    secret="my_secret",
    patch=True,
)

task_id = agent.task_start()
try:
    # OpenAI call tracked automatically
    response = openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Summarise this document..."}],
    )
    agent.ai_response_generated(response.choices[0].message.content[:80], task_id=task_id)
    agent.task_complete("Summary done", task_id=task_id)
except Exception as e:
    agent.task_fail(str(e), task_id=task_id)

---

License

MIT — notilens.com