> ## Documentation Index
> Fetch the complete documentation index at: https://docs.useharmonia.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Harmonia documentation

> Plan, create, approve, publish, and verify startup content with a governed social media agent.

<img src="https://mintcdn.com/harmonia/tzMTcMCHacrkRbcq/brand/harmonia-banner.png?fit=max&auto=format&n=tzMTcMCHacrkRbcq&q=85&s=6a49939e34435b617b6e5af52574fdec" alt="Harmonia — the autonomous social media agent for startups" width="1500" height="500" data-path="brand/harmonia-banner.png" />

Harmonia turns one authorized source into a governed content operation. Its specialist agents find the signal, shape the strategy, produce platform-native work, pause for human authority, and verify every approved external effect.

**Source to signal to proof.** Creative judgment stays agentic. Workflow truth, permissions, cost controls, publishing, receipts, and verification stay deterministic.

<CardGroup cols={2}>
  <Card title="See what Harmonia does" icon="sparkles" href="/product/overview">
    Understand the product promise, operator experience, and governed content loop.
  </Card>

  <Card title="Run the stack" icon="rocket" href="/quickstart">
    Start the web app, ADK worker, Firestore emulator, and Pub/Sub emulator locally.
  </Card>

  <Card title="Follow a content job" icon="diagram-project" href="/pipeline">
    Trace ingestion, analysis, strategy, production, approval, publishing, and proof.
  </Card>

  <Card title="Meet the agent team" icon="people-group" href="/agents/overview">
    Learn how seven bounded specialists research, plan, write, review, present, and assist.
  </Card>

  <Card title="Operate with confidence" icon="shield-check" href="/operations/overview">
    Inspect authority, recovery, observability, replay, deployment, and durable state.
  </Card>

  <Card title="Use exact references" icon="book" href="/reference/overview">
    Look up routes, contracts, states, configuration, storage, budgets, and receipts.
  </Card>
</CardGroup>

## The content loop

1. **Ingest** an authorized YouTube URL, uploaded video, or operator brief.
2. **Understand** the source with Gemini and grounded research.
3. **Strategize** with specialist agents that preserve evidence and uncertainty.
4. **Approve** the exact strategy and consequential effects through deterministic gates.
5. **Produce** reviewed posts, clips, schedules, and content packs.
6. **Publish and verify** through official integrations, idempotent commands, receipts, and fresh provider reads.

## Built for bounded autonomy

* **Agents judge; code authorizes.** Google ADK specialists analyze, plan, write, review, and compose operator surfaces. Application code owns stage transitions, policy, and authority.
* **State survives interruption.** Firestore is workflow truth; Pub/Sub delivers resumable work; failures remain visible and recoverable.
* **Approval is payload-bound.** A conversational “yes” cannot publish by itself. It opens the same exact confirmation flow used by the dashboard.
* **Effects are idempotent and auditable.** Approved actions carry stable identities, produce receipts, and are independently verified.
* **Interfaces stay consistent.** Dashboard, conversational chat, and allow-listed Telegram entry points share the same pipeline and approval boundaries.

## Explore by role

* **Operators:** start with [Product overview](/product/overview), [Pipeline](/pipeline), and [Approval and audit](/approval-and-audit).
* **Developers:** use the [Quickstart](/quickstart), [Platform map](/platform/overview), and [Reference](/reference/overview).
* **Reviewers:** inspect [Architecture](/architecture), [Evaluation](/evaluation), and [Evidence runbook](/evidence-runbook).
* **SREs and maintainers:** follow [Operations](/operations/overview), [Failure recovery](/failure-recovery), and [Observability](/observability).

## Evidence and deployment status

Harmonia documents implemented contracts separately from authenticated deployment evidence. Repository tests and local emulator runs prove local behavior only; provider and cloud claims require captured authenticated evidence. The currently reachable Cloud Run revision is a **web-only preview** and does not prove a deployed worker, Scheduler, Agent Engine invocation, Memory Bank retrieval, or live publish. Use the [evidence runbook](/evidence-runbook) before presenting deployment, model, publishing, or verification claims.

<Note>
  Start with the [quickstart](/quickstart) for local inspection. Use [configuration](/configuration) and [deployment](/deployment) only when you are ready to supply real Google Cloud and platform credentials.
</Note>