# Sentient-X Docs

> Golden paths for building and operating robotics systems with Sentient-X

This document contains the full content of all documentation pages for AI consumption.

---

## Work with agents

**URL:** https://docs.sentientx.io/docs/agents
**Description:** Give agents enough authority to help deeply without giving away accountability

Agents are a first-class interface to Sentient-X. The project connection installs a shared skill
and MCP server; repository agents can also inspect code, run tests, and implement bounded changes.
The best results come from specifying the outcome, boundaries, and proof.

## The four-part brief

| Part | Tell the agent | Example |
|---|---|---|
| Outcome | What should become true | “Make this episode discoverable for training.” |
| Scope | The project, task, member, or failure in bounds | “Dataset `piper-cups`, task revision X.” |
| Boundaries | What requires approval or must not change | “No raw deletion; no physical run; preserve wire types.” |
| Evidence | What proves completion | “Receipt, lineage query, focused test, then relevant gate.” |

## A reliable agent loop

  
  
  
  

## Prompts that scale with experience

  
  

## Keep these decisions human

- widening task or project intent;
- spending outside the stated budget;
- admitting hardware or starting physical motion;
- promoting to production or rolling back a healthy deployment;
- weakening safety, provenance, held-out evaluation, or raw-data guarantees.

Agents can assemble the evidence and recommend an action. The accountable engineer owns these
decisions.

---

## Architecture in one page

**URL:** https://docs.sentientx.io/docs/architecture
**Description:** The four pillars, the handoffs, and the few identities that survive the whole loop

```text
Experience ──▶ Worlds ──▶ Autonomy ──▶ Fleet
    ▲                                      │
    └─ episodes, corrections, outcomes ─┘
```

## The four owners

| Pillar | Owns | Hands off |
|---|---|---|
| Experience | task contracts, episodes, corpora, lineage, access | exact governed inputs |
| Worlds | executable environments, generations, held-out evaluations | promotion evidence |
| Autonomy | training, applications, approvals, improvement decisions | deployable exact graphs |
| Fleet | deployment, station authority, supervision, physical research | executed outcomes and corrections |

## The objects worth remembering

- **Project** — the authorization and product boundary for one robotics application.
- **Task contract reference** — the content identity that says exactly what work means.
- **Episode** — canonical RRD evidence with embodiment, action, provenance, and lineage.
- **Corpus snapshot** — immutable membership used for training or evaluation.
- **Robot application** — the complete digest-pinned deployable graph, not just a model.
- **Qualification and approval** — evidence that the exact graph and task may move forward.
- **Operation reference** — durable ownership of long-running work across disconnects.

## The handoff law

Every gate passes an exact identity plus governed evidence. A receiver does not rebuild another
pillar's facts, infer success from a name, or continue with substitute behavior when authority is
missing.

## Canonical detail

- [Platform model](https://github.com/Sentient-X/sx/blob/main/docs/PLATFORM-MODEL.md)
- [Strategy and golden path](https://github.com/Sentient-X/sx/blob/main/docs/STRATEGY.md)
- [Package ownership](https://github.com/Sentient-X/sx/blob/main/docs/PACKAGES.md)
- [Architecture laws](https://github.com/Sentient-X/sx/blob/main/docs/ABSTRACTIONS.md)

---

## What is real now

**URL:** https://docs.sentientx.io/docs/current-status
**Description:** Find live surfaces and known journey gaps without trusting stale prose

Sentient-X separates architecture, measured implementation, and end-to-end journey evidence. Use
the right source for the question you are asking.

## Three sources of truth

  
  

## Read status precisely

| Label | Meaning |
|---|---|
| Live | The declared route meets the platform admission rule in the repository |
| Planned | The contract is specified but not admitted as a live public tool |
| Deployed | A release manifest contains real digests and infrastructure publishes it |
| Traversed | A named user completed the full journey without hidden intervention |

These labels are not interchangeable. A live verb does not prove a deployed service, and a set of
live verbs does not prove the composed journey.

## Ask an agent for a status brief

```text
For <journey or operation>, compare PLATFORM-MODEL.md, SURFACE-STATUS.generated.md,
USER-JOURNEYS.md, and the current code. Separate designed, live, deployed, and traversed facts.
Cite exact sources, flag contradictions, and do not mutate anything.
```

---

## Data engineer

**URL:** https://docs.sentientx.io/docs/data-engineer
**Description:** Turn immutable robot evidence into governed, reproducible training corpora

Your golden path is **raw capture → canonical RRD episode → catalog lineage → immutable selection
→ lazy training read or export**.

## The path

  
  
  
  

## Give this to your agent

```text
Trace dataset <name> from raw source through canonical RRD registration to its latest immutable
training selection. Cite task refs, episode refs, provenance, transforms, and readiness.
Do not mutate or export. Identify missing lineage and the smallest fail-closed repair.
```

## Review checklist

- One source owns each independent fact; derived indexes and exports can be rebuilt.
- Raw paths are outside every delete, cleanup, and migration action.
- Every episode records the exact embodiment and action interface that produced it.
- Corpus membership is content-addressed before compute begins.
- Training never materializes more than its terminal batch boundary requires.
- Errors identify the failed contract part; no empty string, default, or stale cache hides it.

## Drill down

- [Experience data boundaries](https://github.com/Sentient-X/sx/blob/main/docs/DATA-BOUNDARIES.md)
- [Catalog architecture](https://github.com/Sentient-X/sx/blob/main/experience/data-catalog/docs/ARCHITECTURE.md)
- [Pipeline architecture](https://github.com/Sentient-X/sx/blob/main/experience/data-pipeline/ARCHITECTURE.md)
- [Episode package](https://github.com/Sentient-X/sx/tree/main/packages/sx-episodes)

---

## Deep learning scientist

**URL:** https://docs.sentientx.io/docs/deep-learning-scientist
**Description:** Run fast experiments without detaching models from governed data and promotion evidence

Your golden path is **immutable corpus → training run → comparison → exact application → Worlds
gate → canary**.

## The path

  
  
  
  

## Give this to your agent

```text
Compare the candidate runs for task <task-ref>. Verify that corpus snapshots, recipes,
embodiment domains, and evaluation suites are comparable. Rank hypotheses from governed
evidence, not a single metric. Do not train on held-out evaluation data or promote anything.
Return exact run, artifact, and evidence references plus the next bounded experiment.
```

## Fast without becoming vague

- Change one scientific variable per comparison when practical.
- Prefer a small exact snapshot over an implicit “latest” dataset.
- Keep arrays and padded tensors at the terminal model boundary.
- Inspect per-domain and failure-slice metrics, not only the global mean.
- Promote an application identity, never a local checkpoint path.
- Reattach to durable operations; a disconnected client is not a failed experiment.

```bash
OPERATION=$(sx --project "$PROJECT_ID" worlds evaluate policy:pick-mug-v12)
sx watch "$OPERATION"
```

## Drill down

- [Autonomy model](https://github.com/Sentient-X/sx/blob/main/docs/PLATFORM-MODEL.md#autonomy)
- [Training decisions](https://github.com/Sentient-X/sx/blob/main/autonomy/train/docs/DECISIONS.md)
- [Waddle](https://github.com/Sentient-X/sx/tree/main/autonomy/waddle)
- [Typed SDK golden path](https://github.com/Sentient-X/sx/blob/main/packages/sx/tests/test_typed_golden_path.py)

---

## The golden path

**URL:** https://docs.sentientx.io/docs/golden-path
**Description:** One task, one lineage, and one evidence chain from experience to fleet

The golden path is the shortest complete robotics story: define one task, gather governed
experience, evaluate and improve an exact application, deploy it under supervision, and return
executed outcomes to the catalog.

## The lifecycle

  
  
  
  
  

## Let an agent drive the routine work

After `sx project connect`, give your coding agent an accountable goal and a budgeted boundary:

```text
Move task <task-ref> toward a supervised canary using the Sentient-X MCP tools.
First inspect the current resources and cite every exact reference. Use the canonical
Experience → Worlds → Autonomy → Fleet path. Do not widen the task, spend budget,
start physical motion, or promote beyond canary without asking. Reattach to durable
operations and report success only from terminal receipts and governed evidence.
```

The agent should return a trail you can audit:

- exact task, corpus, application, qualification, approval, and deployment references;
- durable operation references and terminal states;
- evaluation or RRD evidence;
- every refusal and the next safe recovery action.

## The same model in Python

The SDK keeps pillar ownership visible and returns typed durable operations.

```python
from uuid import UUID

from sx_worlds import WorldEvaluation

async def evaluate(project_id: UUID) -> WorldEvaluation:
    async with sx.connect(project_id) as client:
        policy = await client.autonomy.policy("policy:pick-mug-v12")
        operation = await client.worlds.evaluate(policy)
        print(operation.ref)  # save this; disconnecting does not cancel work
        return await operation
```

For the complete typed capture-to-deploy contract, read the executable
[golden-path test](https://github.com/Sentient-X/sx/blob/main/packages/sx/tests/test_typed_golden_path.py).

## One promotion rule

Promote evidence, not confidence. A result moves forward only when its owning gate can prove the
exact identity that passed. Keep exploration flexible inside a stage; keep handoffs strict.

---

## Build the loop

**URL:** https://docs.sentientx.io/docs
**Description:** Simple, agent-first golden paths for serious robotics work

Sentient-X turns a task into governed experience, an evaluated robot application, and a
supervised deployment. The loop closes when fleet outcomes become the next round of experience.

```text
Experience → Worlds → Autonomy → Fleet → Experience
```

  
  
  

## Choose your altitude

The same system should work for an intern shipping a focused fix and a senior engineer tracing a
cross-pillar failure. Pick the path closest to the result you own.

  
  
  

## The operating idea

You do not need the whole repository in your head. Start from the outcome, find the owning pillar,
cross one typed boundary at a time, and keep the task reference and evidence attached. Use an agent
to map unfamiliar code and operate routine tools; use tests, receipts, and generated status as the
proof.

---

## Intern

**URL:** https://docs.sentientx.io/docs/intern
**Description:** Learn the system by shipping one small, well-proven vertical change

You are not expected to memorize the monorepo. Your first job is to build a dependable map and
make one outcome measurably better.

## Four things to remember

| Name | Meaning |
|---|---|
| Experience | What the work means and what robots have done |
| Worlds | Where behavior is generated and evaluated before physical deployment |
| Autonomy | How governed experience becomes an ability to act |
| Fleet | How exact applications run on robots under supervision |

The task contract connects all four. Tests, receipts, and evidence tell you whether a change is
real.

## Your first-PR golden path

  
  
  
  
  

## Copy this prompt

```text
Help me understand and implement <issue>. Read CLAUDE.md, docs/DESIGN-POLICY.md,
and the affected member guidance. First explain the real request-to-effect path with file
citations and the focused test. Then make the smallest complete change, preserving safety,
lineage, and typed failures. Run focused checks and show me how the evidence proves the outcome.
Do not touch unrelated changes.
```

## Stop and ask when

- an action could delete or rewrite raw robot data;
- a test would train on or relax held-out evaluation data;
- a change affects physical motion, safe-stop, control authority, or deployment promotion;
- the requested fix needs a new public contract or changes persisted wire values;
- the repository and the issue disagree about what should be true.

## Good next reads

- [Start here](/docs/start-here)
- [Design policy](https://github.com/Sentient-X/sx/blob/main/docs/DESIGN-POLICY.md)
- [Repository guide](https://github.com/Sentient-X/sx/blob/main/CLAUDE.md)
- [Agent collaboration](/docs/agents)

---

## Robotics engineer

**URL:** https://docs.sentientx.io/docs/robotics-engineer
**Description:** Bring an exact embodiment and task to a supervised canary without losing safety or lineage

Your golden path is **task contract → embodiment → Worlds evidence → qualified application →
supervised canary**.

## The path

  
  
  
  

## Give this to your agent

```text
Inspect the path for task <task-ref> on embodiment <embodiment-ref>. Resolve exact interfaces,
current Worlds evidence, application qualification, approvals, deployment group, and Fleet
readiness. Do not start physical motion. Return blockers in safety order and the smallest
evidence-producing next action, with exact references.
```

## Failure questions senior engineers ask

- Is the station still the sole actuator authority?
- Does the served policy identity match the catalog application and action interface exactly?
- Are commands expiring and sequenced, or can stale intent replay?
- Does an unanswered escalation terminate in safe-stop?
- Did the robot execute every action written as training evidence?

## Drill down

- [Platform model: Worlds and Fleet](https://github.com/Sentient-X/sx/blob/main/docs/PLATFORM-MODEL.md)
- [Fleet station runtime plan](https://github.com/Sentient-X/sx/blob/main/docs/plans/fleet-station-runtime-2026-07.md)
- [Hosted research runbook](https://github.com/Sentient-X/sx/blob/main/fleet/supervisors/docs/hosted-research.md)
- [Embodiment contract](https://github.com/Sentient-X/sx/blob/main/docs/EMBODIMENT-CONTRACT.md)

---

## Start here

**URL:** https://docs.sentientx.io/docs/start-here
**Description:** Reach a useful, verified result without learning the whole platform first

Pick the door that matches your job today.

  

## Your first useful result

  
  
  

## Two commands worth remembering

```bash
make check      # repository-wide validation
sx watch REF    # reattach to durable platform work
```

## Next

- Follow the [platform golden path](/docs/golden-path).
- Learn the [agent collaboration pattern](/docs/agents).
- Choose a [role-specific path](/docs/robotics-engineer).

---

## Links

- [GitHub](https://github.com/Sentient-X/sx)