Pulse

Pulse is a Phase 1 MVP for an AI operations cockpit. It demonstrates an end-to-end incident workflow: telemetry ingestion, incident creation, specialist investigation, human approval before remediation, mock execution, postmortem generation, and memory graph updates.

The current version is intentionally lightweight:

• FastAPI backend
• SQLite storage
• Vanilla JS dashboard
• Deterministic specialist agents for demo flow
• CLI helpers for seeding and triggering incidents

Current Demo Scope

Pulse currently proves these pieces:

• Event ingestion opens an incident on high latency
• Supervisor orchestration calls specialist agents
• Metrics, logs, context, risk, and remediation findings are merged
• Remediation is blocked until approval
• Executor resolves the incident only after approval
• Postmortem data is generated after execution
• Memory graph is updated after resolution

This is a workflow MVP, not a full autonomous AI system yet.

Architecture

Dashboard UI (Vanilla JS)
        |
        v
FastAPI App
  |- Landing Page
  |- Dashboard Page
  |- API Routes
        |
        v
Event Ingestion
        |
        v
SQLite Store
  |- Events
  |- Incidents
  |- Approvals
  |- Memory Graph
        |
        v
Supervisor
  |- Metrics Agent
  |- Logs Agent
  |- Context Agent
  |- Risk Agent
  |- Remediation Agent
        |
        v
Approval Gate
        |
        v
Executor
        |
        v
Postmortem + Memory Update

Project Layout

Pulse/
  pulse/
    agents/
    api/
    approvals/
    context/
    executor/
    observability/
    orchestration/
    models.py
    store.py
    cli.py
  tests/
  Makefile
  pyproject.toml

Run The Dashboard

From the project root:

python3 -m pip install -e '.[test]'
make dev

Then open:

• http://127.0.0.1:8000/
• http://127.0.0.1:8000/dashboard

Recommended demo flow:

1. Click Seed Demo
2. Click Trigger Incident
3. Review the incident card, timeline, findings, and approval panel
4. Click Approve
5. Click Execute
6. Show the postmortem preview and memory updates

Run The CLI

Seed the demo data:

make seed

Trigger an incident from CLI:

make trigger-incident

Run tests:

make test

You can also use the module entrypoints directly:

python3 -m pulse.cli seed
python3 -m pulse.cli trigger-incident

Key Routes

UI routes:

• GET /
• GET /dashboard
• GET /memory

Core API routes:

• POST /events/ingest
• GET /incidents
• GET /incidents/{id}
• POST /incidents/{id}/investigate
• GET /approvals
• POST /approvals/{id}/approve
• POST /approvals/{id}/reject
• POST /executor/run/{approval_id}
• GET /memory/graph
• GET /demo/seed
• POST /demo/trigger-incident

How It Works Today

The current specialist agents are deterministic Python functions, not live LLM-backed agents.

• Metrics Agent compares latency values
• Logs Agent groups repeated log summaries
• Context Agent builds a compact context pack
• Risk Agent marks remediation as approval-gated
• Remediation Agent returns a safe mock action

This keeps the MVP fast, testable, and easy to demo while preserving the intended architecture.

Future Scope

Short-term upgrades:

• Replace deterministic log/context analysis with Ollama or model-backed reasoning
• Add multiple incident scenarios instead of a single seeded path
• Improve memory graph usefulness with similarity search
• Move startup hooks to FastAPI lifespan
• Add richer dashboard visualizations for metrics and approvals

Longer-term scope:

• Real observability ingestion
• Background workers
• Multi-service incident correlation
• Safer remediation playbooks
• Deployable hosted frontend/backend split