Skip to content

PULSE — Autonomous Virtual Employee

PULSE is the flagship feature of buddy-ai v2.1. It is a fully autonomous virtual team member — not a chatbot you prompt, but an AI employee that works through your task queue, learns from your documents, attends your meetings, and reports back, all without being managed.

Quick Start

pip install buddy-ai
buddy pulse start
# Opens http://localhost:8888

That's it. The web UI walks you through creating your virtual employee.

What PULSE Does Autonomously

Once created, a PULSE employee operates independently:

Behaviour How
Works tasks Picks the highest-priority todo task, executes it, and creates a real file (.py, .md, .sql, etc.) saved to the employee's workspace — every 30 seconds
Remembers Automatically extracts key facts from every conversation and persists them across server restarts — names, preferences, deadlines, project context
Sends notifications Notifies you every time it completes work with a summary of what it created
Daily standup Generates a first-person standup report once per day — what it worked on, what's next, any observations
Suggests tasks Proactively proposes 3 high-value tasks based on its role and knowledge — once per day
Activity feed Logs every autonomous action with timestamp — visible live on the Dashboard
Learns from documents Reads PDFs, crawls entire websites (BFS, up to 20 pages), and retains knowledge permanently
Attends meetings Processes transcripts, extracts decisions and action items, creates tasks from its own action items
Answers from knowledge Every chat response uses everything it has learned — KT sessions + memories injected as context

Web UI

buddy pulse start
buddy pulse start --port 3000 --no-browser

Dashboard

The command centre for your virtual employee:

  • Activity Feed — live log of every action PULSE takes (tasks started/done, KT learned, standups, suggestions)
  • Currently Working On — shows in-progress tasks with a live pulsing indicator
  • Daily Standup — generate on demand or let it auto-generate once per day
  • PULSE Suggests — PULSE proposes proactive tasks based on its role and knowledge; add them with one click
  • Knowledge Domains — all topics PULSE has been trained on

Task Board

Kanban board with full autonomous execution:

  • Add a task → PULSE picks it up within 30 seconds
  • Blue pulsing dot on a card means PULSE is actively working on it
  • When done, the card shows a 📄 filename.py link to the generated file in the Workspace
  • Auto / Manual toggle — disable auto-work if you want to control task execution yourself
  • Board auto-refreshes every 15 seconds

Workspace (new)

Every task PULSE completes creates a real file, saved to ~/.pulse_data/workspaces/{employee_id}/:

Task type File created
Write a Python script .py
Write SQL query .sql
Create a report / document .md
Generate a config .yaml / .json
Write a shell script .sh
Create a webpage .html

The Workspace page in the sidebar shows all files PULSE has created:

  • Browse and filter by file type
  • View file contents with syntax highlighting (code) or markdown rendering
  • Download any file directly from the browser
  • Delete files you no longer need

Memory (new)

PULSE maintains two persistent memory layers per employee:

Long-term Memories - After every conversation turn, PULSE runs a background extraction to identify facts worth remembering: user preferences, project context, deadlines, colleague names, key decisions - Up to 100 facts stored per employee - Each fact is injected into future conversations: PULSE greets you by context, not by blank slate - View, search, and selectively forget individual facts from the Memory page

Conversation History - Last 30 messages are kept as a rolling context window - Injected verbatim into every prompt so PULSE remembers what was said earlier in the session — and in past sessions - Survives server restarts

Both layers are persisted to ~/.pulse_data/employees.json and restored on startup.

KT Center

Teach PULSE from any source:

  • Document / PDF — upload or provide a file path; text is extracted with pypdf
  • URL — PULSE crawls the entire website (BFS up to 20 pages, 3 hops deep, same domain)
  • Live KT — real-time interactive session where PULSE asks Socratic questions until it reaches ≥82% confidence

Meeting Room

Paste any meeting transcript. PULSE extracts:

  • Summary
  • Decisions made
  • Action items with owners
  • Its own action items are automatically added to the Task Board

Chat

Streaming WebSocket conversation. Every response uses: - Full KT knowledge base as context - Long-term memories about you - Recent conversation history

Ask about any document it has read and it answers as a team member who was there — and remembers what you told it last week.

Notifications

Bell icon in the sidebar with unread count. PULSE sends you a notification for:

  • Every task it completes (with file created + summary)
  • Daily standup reports
  • Proactive task suggestions

Settings

Configure LLM model and API key directly in the UI — no environment variables needed. Supports:

  • OpenAI — 22 models including GPT-4.1, GPT-4o, o3, o4-mini, o1
  • Anthropic — Claude Opus, Sonnet, Haiku
  • Google — Gemini 1.5 Pro, 2.0 Flash
  • Groq — Llama 3.3, Mixtral
  • Ollama — any local model

Changes apply immediately to all active employees.

Python API

Create an employee

from buddy.pulse import PulseEmployee, EmployeeProfile, KTSourceType
from buddy.models.openai import OpenAIChat

pulse = PulseEmployee(
    employee_profile=EmployeeProfile(
        full_name="Priya Sharma",
        role="Senior Backend Engineer",
        department="Engineering",
        skills=["Python", "FastAPI", "PostgreSQL"],
        timezone="Asia/Kolkata",
        company_name="Acme Corp",
    ),
    model=OpenAIChat(id="gpt-4o"),
)

print(pulse.introduce_yourself())

Knowledge Transfer

# From a PDF
summary = pulse.take_kt(
    source="docs/payments_architecture.pdf",
    source_type=KTSourceType.DOCUMENT,
    session_name="Payments KT",
    knowledge_giver="Arjun Nair",
)

# From a URL (crawls the entire site)
summary = pulse.take_kt(
    source="https://docs.yourproduct.com",
    source_type=KTSourceType.URL,
    session_name="Product Docs KT",
    knowledge_giver="Docs Site",
)

print(f"Confidence: {summary.confidence_score:.0%}")
print(f"Mental model: {summary.mental_model}")

Live KT (interactive)

session = pulse.start_live_kt(
    session_name="Auth Service KT",
    knowledge_giver="Arjun Nair",
    source_type=KTSourceType.HUMAN_CHAT,
)

turn = session.human_explains("Our auth uses JWT with 1-hr expiry...")
print(turn.pulse_message)  # PULSE asks targeted questions

turn = session.human_answers({"Q1": "User must re-login", "Q2": "Yes, Redis blocklist"})
print(f"Confidence: {turn.confidence_score:.0%}")

summary = pulse.finalize_kt_session(session)

Meetings

from buddy.pulse import MeetingPlatform

notes = pulse.attend_meeting(
    transcript="Arjun: Complete the Razorpay migration by EOD. Priya: I'll take the refund flow.",
    participants=["Arjun Nair", "Priya Sharma"],
    platform=MeetingPlatform.ZOOM,
    title="Sprint Sync",
)

print(notes.format_summary())
# Action items assigned to PULSE are auto-added to its task list

Tasks

task = pulse.receive_task(
    title="Implement Razorpay refund flow",
    description="Handle refund_processed webhook events...",
    assigned_by="Arjun Nair",
    priority="high",
    deadline="2026-06-27",
)

task.start()
task.add_note("Reviewed Razorpay refund webhook docs")

update = pulse.report_status(task_id=task.task_id)
print(update.format_standup())

Data Persistence

All employee data is saved to ~/.pulse_data/employees.json on every change:

  • Employee profile
  • All KT session summaries and mental models
  • Task list with statuses, notes, and work output
  • Conversation history (last 30 messages)
  • Long-term memory facts (up to 100 per employee)
  • LLM configuration

Workspace files are saved to ~/.pulse_data/workspaces/{employee_id}/.

On buddy pulse start, everything is automatically restored — no re-onboarding needed.

REST API Reference

When running buddy pulse start, the full API is available at http://localhost:8888/api/docs.

Employee

Method Endpoint Description
POST /api/pulse/employees Create a PULSE employee
GET /api/pulse/employees/{id} Get employee profile + status
PUT /api/pulse/employees/{id} Update employee profile

Knowledge Transfer

Method Endpoint Description
POST /api/pulse/{id}/kt/async KT from document, PDF, or URL
POST /api/pulse/{id}/kt/live Start a live KT session
POST /api/pulse/{id}/kt/{sid}/explain Human explains a chunk
POST /api/pulse/{id}/kt/{sid}/answer Human answers questions
POST /api/pulse/{id}/kt/{sid}/commit Finalise the session
GET /api/pulse/{id}/kt List all KT sessions

Tasks

Method Endpoint Description
POST /api/pulse/{id}/tasks Assign a task
GET /api/pulse/{id}/tasks List all tasks
PUT /api/pulse/{id}/tasks/{tid} Update status or add a note
GET /api/pulse/{id}/tasks/{tid}/status-update Get AI status report

Workspace (new)

Method Endpoint Description
GET /api/pulse/{id}/workspace List all files PULSE created
GET /api/pulse/{id}/workspace/{filename} Get file content
DELETE /api/pulse/{id}/workspace/{filename} Delete a workspace file

Memory (new)

Method Endpoint Description
GET /api/pulse/{id}/memory Get all memories + conversation history
DELETE /api/pulse/{id}/memory/{memory_id} Forget a single fact
DELETE /api/pulse/{id}/memory Clear all memory
DELETE /api/pulse/{id}/memory/history/clear Clear conversation history only

Meetings

Method Endpoint Description
POST /api/pulse/{id}/meetings Process a meeting transcript

Chat

Method Endpoint Description
POST /api/pulse/{id}/chat Single-turn chat (KT + memory aware)
WS /api/pulse/{id}/chat/stream Streaming WebSocket chat

Activity & Notifications

Method Endpoint Description
GET /api/pulse/{id}/activity Get activity log
GET /api/pulse/{id}/notifications Get notifications with unread count
POST /api/pulse/{id}/notifications/read-all Mark all as read

Standup & Suggestions

Method Endpoint Description
POST /api/pulse/{id}/standup Generate daily standup
POST /api/pulse/{id}/suggest-tasks Get proactive task suggestions

Settings

Method Endpoint Description
GET /api/pulse/settings/llm Get LLM configuration
POST /api/pulse/settings/llm Save LLM configuration
POST /api/pulse/settings/llm/test Test LLM connection
GET /api/pulse/settings/auto-work Get auto-work status
POST /api/pulse/settings/auto-work Enable / disable auto-work

Architecture

buddy/pulse/
├── employee.py      # PulseEmployee (extends Agent)
├── identity.py      # EmployeeProfile, WorkingHours, WorkStyle
├── kt.py            # KTSession, KTManager, KTSourceType, KTSummary — PDF + BFS URL crawler
├── meeting.py       # MeetingParticipant, MeetingNotes, ActionItem, TranscriptProcessor
├── work.py          # WorkItem, TaskManager, WorkCalendar, StatusUpdate
├── comms.py         # CommunicationHub (Slack, Gmail, Teams, Zoom)
├── memory.py        # ProfessionalMemory
├── onboarding.py    # OnboardingWorkflow
├── feedback.py      # FeedbackSystem, PerformanceTracker
├── router.py        # FastAPI endpoints + autonomous background worker + memory system
├── app.py           # PulseApp (serves React UI + API, starts auto worker on startup)
└── __init__.py      # Public exports

The autonomous background worker runs inside the FastAPI event loop:

  1. Every 30 seconds, checks each employee's todo queue
  2. Picks the highest-priority task, infers the output format (.py, .md, .sql, etc.)
  3. Runs the LLM in a thread pool (non-blocking) with a KT-aware prompt
  4. Saves the generated content as a real file in ~/.pulse_data/workspaces/{employee_id}/
  5. Records the filename in the task card — visible and downloadable from the UI
  6. Sends a notification and logs the event to the activity feed
  7. Once the queue is empty, generates the daily standup (once per day)
  8. Suggests proactive tasks (once per day)

The memory system runs as a background asyncio.create_task after every chat message:

  1. Sends a focused extraction prompt to the LLM asking for 0–3 facts worth remembering
  2. Deduplicates against existing facts (no repeated entries)
  3. Appends new facts to ~/.pulse_data/employees.json
  4. Next conversation turn injects the accumulated facts + recent history into the prompt

CLI

# Start the server
buddy pulse start
buddy pulse start --port 3000 --no-browser

# Create an employee config interactively
buddy pulse create --name "Priya Sharma" --role "Backend Engineer"

# Run a KT session from the CLI
buddy pulse kt docs/architecture.pdf --session "Architecture KT" --from "Arjun"

# Check module status
buddy pulse status

Web UI

buddy pulse start
buddy pulse start --port 3000 --no-browser

Dashboard

The command centre for your virtual employee:

  • Activity Feed — live log of every action PULSE takes (tasks started/done, KT learned, standups, suggestions)
  • Currently Working On — shows in-progress tasks with a live pulsing indicator
  • Daily Standup — generate on demand or let it auto-generate once per day
  • PULSE Suggests — PULSE proposes proactive tasks based on its role and knowledge; add them with one click
  • Knowledge Domains — all topics PULSE has been trained on

Task Board

Kanban board with full autonomous execution:

  • Add a task → PULSE picks it up within 30 seconds
  • Blue pulsing dot on a card means PULSE is actively working on it
  • When done, the full AI-generated work output appears in the card
  • Auto / Manual toggle — disable auto-work if you want to control task execution yourself
  • Board auto-refreshes every 15 seconds

KT Center

Teach PULSE from any source:

  • Document / PDF — upload or provide a file path; text is extracted with pypdf
  • URL — PULSE crawls the entire website (BFS up to 20 pages, 3 hops deep, same domain)
  • Live KT — real-time interactive session where PULSE asks Socratic questions until it reaches ≥82% confidence

Meeting Room

Paste any meeting transcript. PULSE extracts:

  • Summary
  • Decisions made
  • Action items with owners
  • Its own action items are automatically added to the Task Board

Chat

Streaming WebSocket conversation. Every response uses the full KT knowledge base as context — ask about any document it has read and it answers as a team member who was there.

Notifications

Bell icon in the sidebar with unread count. PULSE sends you a notification for:

  • Every task it completes (with work summary)
  • Daily standup reports
  • Proactive task suggestions

Settings

Configure LLM model and API key directly in the UI — no environment variables needed. Supports:

  • OpenAI — 22 models including GPT-4.1, GPT-4o, o3, o4-mini, o1
  • Anthropic — Claude Opus, Sonnet, Haiku
  • Google — Gemini 1.5 Pro, 2.0 Flash
  • Groq — Llama 3.3, Mixtral
  • Ollama — any local model

Changes apply immediately to all active employees.

Python API

Create an employee

from buddy.pulse import PulseEmployee, EmployeeProfile, KTSourceType
from buddy.models.openai import OpenAIChat

pulse = PulseEmployee(
    employee_profile=EmployeeProfile(
        full_name="Priya Sharma",
        role="Senior Backend Engineer",
        department="Engineering",
        skills=["Python", "FastAPI", "PostgreSQL"],
        timezone="Asia/Kolkata",
        company_name="Acme Corp",
    ),
    model=OpenAIChat(id="gpt-4o"),
)

print(pulse.introduce_yourself())

Knowledge Transfer

# From a PDF
summary = pulse.take_kt(
    source="docs/payments_architecture.pdf",
    source_type=KTSourceType.DOCUMENT,
    session_name="Payments KT",
    knowledge_giver="Arjun Nair",
)

# From a URL (crawls the entire site)
summary = pulse.take_kt(
    source="https://docs.yourproduct.com",
    source_type=KTSourceType.URL,
    session_name="Product Docs KT",
    knowledge_giver="Docs Site",
)

print(f"Confidence: {summary.confidence_score:.0%}")
print(f"Mental model: {summary.mental_model}")

Live KT (interactive)

session = pulse.start_live_kt(
    session_name="Auth Service KT",
    knowledge_giver="Arjun Nair",
    source_type=KTSourceType.HUMAN_CHAT,
)

turn = session.human_explains("Our auth uses JWT with 1-hr expiry...")
print(turn.pulse_message)  # PULSE asks targeted questions

turn = session.human_answers({"Q1": "User must re-login", "Q2": "Yes, Redis blocklist"})
print(f"Confidence: {turn.confidence_score:.0%}")

summary = pulse.finalize_kt_session(session)

Meetings

from buddy.pulse import MeetingPlatform

notes = pulse.attend_meeting(
    transcript="Arjun: Complete the Razorpay migration by EOD. Priya: I'll take the refund flow.",
    participants=["Arjun Nair", "Priya Sharma"],
    platform=MeetingPlatform.ZOOM,
    title="Sprint Sync",
)

print(notes.format_summary())
# Action items assigned to PULSE are auto-added to its task list

Tasks

task = pulse.receive_task(
    title="Implement Razorpay refund flow",
    description="Handle refund_processed webhook events...",
    assigned_by="Arjun Nair",
    priority="high",
    deadline="2026-06-27",
)

task.start()
task.add_note("Reviewed Razorpay refund webhook docs")

update = pulse.report_status(task_id=task.task_id)
print(update.format_standup())

Data Persistence

All employee data is saved to ~/.pulse_data/employees.json on every change:

  • Employee profile
  • All KT session summaries and mental models
  • Task list with statuses, notes, and work output
  • LLM configuration

On buddy pulse start, everything is automatically restored — no re-onboarding needed.

REST API Reference

When running buddy pulse start, the full API is available at http://localhost:8888/api/docs.

Employee

Method Endpoint Description
POST /api/pulse/employees Create a PULSE employee
GET /api/pulse/employees/{id} Get employee profile + status
PUT /api/pulse/employees/{id} Update employee profile

Knowledge Transfer

Method Endpoint Description
POST /api/pulse/{id}/kt/async KT from document, PDF, or URL
POST /api/pulse/{id}/kt/live Start a live KT session
POST /api/pulse/{id}/kt/{sid}/explain Human explains a chunk
POST /api/pulse/{id}/kt/{sid}/answer Human answers questions
POST /api/pulse/{id}/kt/{sid}/commit Finalise the session
GET /api/pulse/{id}/kt List all KT sessions

Tasks

Method Endpoint Description
POST /api/pulse/{id}/tasks Assign a task
GET /api/pulse/{id}/tasks List all tasks
PUT /api/pulse/{id}/tasks/{tid} Update status or add a note
GET /api/pulse/{id}/tasks/{tid}/status-update Get AI status report

Meetings

Method Endpoint Description
POST /api/pulse/{id}/meetings Process a meeting transcript

Chat

Method Endpoint Description
POST /api/pulse/{id}/chat Single-turn chat (KT-aware)
WS /api/pulse/{id}/chat/stream Streaming WebSocket chat

Activity & Notifications

Method Endpoint Description
GET /api/pulse/{id}/activity Get activity log
GET /api/pulse/{id}/notifications Get notifications with unread count
POST /api/pulse/{id}/notifications/read-all Mark all as read

Standup & Suggestions

Method Endpoint Description
POST /api/pulse/{id}/standup Generate daily standup
POST /api/pulse/{id}/suggest-tasks Get proactive task suggestions

Settings

Method Endpoint Description
GET /api/pulse/settings/llm Get LLM configuration
POST /api/pulse/settings/llm Save LLM configuration
POST /api/pulse/settings/llm/test Test LLM connection
GET /api/pulse/settings/auto-work Get auto-work status
POST /api/pulse/settings/auto-work Enable / disable auto-work

Architecture

buddy/pulse/
├── employee.py      # PulseEmployee (extends Agent)
├── identity.py      # EmployeeProfile, WorkingHours, WorkStyle
├── kt.py            # KTSession, KTManager, KTSourceType, KTSummary — includes PDF + URL crawler
├── meeting.py       # MeetingParticipant, MeetingNotes, ActionItem, TranscriptProcessor
├── work.py          # WorkItem, TaskManager, WorkCalendar, StatusUpdate
├── comms.py         # CommunicationHub (Slack, Gmail, Teams, Zoom)
├── memory.py        # ProfessionalMemory
├── onboarding.py    # OnboardingWorkflow
├── feedback.py      # FeedbackSystem, PerformanceTracker
├── router.py        # FastAPI endpoints + autonomous background worker
├── app.py           # PulseApp (serves React UI + API, starts auto worker on startup)
└── __init__.py      # Public exports

The autonomous background worker runs inside the FastAPI event loop:

  1. Every 30 seconds, checks each employee's todo queue
  2. Picks the highest-priority task, runs it through the LLM (in a thread pool, non-blocking)
  3. Stores the full work output in the task's progress_notes
  4. Sends a notification and logs the event to the activity feed
  5. Once the queue is empty, generates the daily standup (once per day)
  6. Suggests proactive tasks (once per day)

CLI

# Start the server
buddy pulse start
buddy pulse start --port 3000 --no-browser

# Create an employee config interactively
buddy pulse create --name "Priya Sharma" --role "Backend Engineer"

# Run a KT session from the CLI
buddy pulse kt docs/architecture.pdf --session "Architecture KT" --from "Arjun"

# Check module status
buddy pulse status