# Paploo by Spadina Labs

> Paploo is an AI assistant for teams that answers with sources and acts with approval. It works in team chat with a company's own tools and data, uses each person's own permissions, keeps credentials away from the AI model, and runs in the customer's cloud or data center with the model provider they choose.

Paploo is made by Spadina Labs (https://spadinalabs.com). It is in early access: teams can book a pilot at https://spadinalabs.com/#pilot.

## What Paploo does

- Answers with sources: every claim links to the log, deploy, ticket or document it came from, read at the moment of the question. When Paploo can't confirm something, it says so and names what it would check next.
- Changes need approval: anything that changes a system waits for a person or a rule the team set. Every action records who asked, what ran and who approved it.
- Each person's own access: Paploo acts with the permissions of the person asking, not a shared bot account. What someone can't open, Paploo doesn't show, name or use in their answer.
- Credentials never reach the model: credentials are encrypted at rest and added by a separate proxy on the way out. Requests reach only allowed hosts.
- Shared workstreams: an incident or project becomes one page with stages, owners, evidence and the next step.
- Handoffs: at shift change Paploo writes what is done, open and to watch, and re-checks every link against the next person's access.
- Company map: services, configs, changes, teams, people and past incidents, connected and drawn per person.
- Lessons and playbooks: postmortems become lessons; a person approves them; the next run starts from them. Lessons keep the access of their source.
- Automations: scheduled or event-driven work such as incident first response, CI triage and release checks, which pings a person only when a decision is needed.
- Health checks: Paploo tests its own connections, tokens and budget on a schedule and flags problems before they break a run.

## Deployment and models

- Runs in your own cloud account (AWS, Google Cloud or Azure) or data center, on Kubernetes with Helm or on a single server with Docker Compose.
- Model providers: Anthropic, OpenAI, AWS Bedrock, Google Vertex AI, any OpenAI-compatible API, or self-hosted models on vLLM or Ollama. Switching providers doesn't change connections, permissions or playbooks.
- Works with Slack, GitLab, GitHub, Jira, Confluence, Kubernetes, SSH hosts and any MCP server.
- Admin controls: single sign-on (Okta, Google or any OIDC provider), roles, approvals, audit trail and usage limits. Paploo doesn't train models on customer data.

## Use cases

- Incident response: cause, rollback and evidence in the thread, then the rest of the incident as one workstream.
- On-call handoff: done, open and watch items with links, re-checked for the next person's access.
- New engineer onboarding: ask the company map who owns what, with every answer linked to its record.
- Customer escalation: support and engineering share one workstream; support sees the customer update, not the logs.
- Code review: a first review in minutes, with comments that cite the line, the rule and past lessons.
- Release checks: run on schedule and learn from what the last release missed.

## Team

- Dagm Fekadu, co-founder: https://www.linkedin.com/in/dagmfekadu/
- Ilya Dashkovsky, co-founder: https://www.linkedin.com/in/dashkovsky/

## Links

- Website: https://spadinalabs.com/
- Book a pilot: https://spadinalabs.com/#pilot