Skip to main content

Trust Mesh walkthrough

Prove & govern an agent run
— in 90 seconds.

Capability token → signed receipt → deterministic replay.

  • Launch: deterministic slice.
  • Guard: OPA→Wasm policies with budgets.
  • Verify: WebCrypto checks the receipt.
Prefer step-by-step? Open the tutorial

Press run → watch receipts appear → verify with the CLI → replay.

  • Capability TokenIdle

    Mint when demo runs

    Start the demo to mint a token.

  • SLO gateIdle

    SLO gate: waiting for run

    Run Hello Fleet to evaluate queue latency.

  • Guardrail decisionsIdle

    Awaiting verdict

    Flip guardrail mode above to fire decisions.

  • C2PA manifestIdle

    Waiting for manifest

    Receipt generates after the run.

  • Replay tokenIdle

    Awaiting replay diff

    Load the Replay & diff section.

  • Capability Token
  • SLO gate
  • Guardrail decisions
  • C2PA manifest
  • Replay token

Legend: Idle waits for a run, Pending is in progress, OK ✓ passed its check, and Drift/Error needs a closer look.

Receipt → Policy decision → Signed artifact → Replay token.

  1. 1

    Copy the run_id

    Every receipt + replay references the same ID.

    Launch the demo ↑

    Use this ID in steps 2 & 3 below.

  2. 2

    Verify the receipt

    Jump straight to the manifest checker with that same ID.

  3. 3

    Replay & diff

    Confirm deterministic replay, then link evidence back to Hello Fleet.

Fleet health & SLOs

This mirrors the landing metrics in plain English: are runs keeping up, and are we staying within the latency + drift budgets that matter?

Delivery guarantees

Hello Fleet demo · Rolling 1h
How we measure this
p95 queue lag
Budget adherence

Mesh vitals

Queued
Running
Succeeded
Failed
Token & cost budgets (OpenTelemetry GenAI)View JSON ↗
Budget used (tokens)
Explain how we compute budget usage

Budget = daily cap for the demo environment. tokens_pct maps to OTel gen_ai.client.token.usage; cost_pct maps to OTel gen_ai.cost.usd.

Names align with the OpenTelemetry GenAI semantic conventions.

Source: /metrics/demo.json (refreshes ~5s). Math + window documented in GenAI telemetry coverage →.

Budget used (USD)
Explain how we compute budget usage

Budget = daily cap for the demo environment. tokens_pct maps to OTel gen_ai.client.token.usage; cost_pct maps to OTel gen_ai.cost.usd.

Names align with the OpenTelemetry GenAI semantic conventions.

Source: /metrics/demo.json (refreshes ~5s). Math + window documented in GenAI telemetry coverage →.

Updated —

01 — Launch

Click once to run the canonical LangGraph slice—this is where you watch capability tokens mint and policy verdicts fire in real time.

Demo run

Run ID Start run to mint
Ready

Baseline shows raw outputs; Guardrails shows post-policy “after”.

  1. plannerPendingplanner – Pending
  2. researcherPendingresearcher – Pending
  3. engineerPendingengineer – Pending
  4. editorPendingeditor – Pending
  5. policyPendingpolicy – Pending
  6. receiptPendingreceipt – Pending

02 — Verify the receipt

Prove the output hasn’t been tampered with: WebCrypto validates the demo C2PA-style manifest—no downloads, no uploads, and the verdict stays local.

Receipt

C2PA-style manifest (demo)

WebCrypto runs locally—no files leave the browser.

SCITT / Biscuit exports optional; see Status & Acceptance.

Download manifest
trust.capability_token_id
Launch the demo ↑
trust.attestation_id
Launch the demo ↑
run_id
hf_demo
View manifest JSON
{}

Tip: the landing page’s “Receipt preview” mirrors this chain so `/` and `/demo` stay in sync. Latest run ID: Launch the demo ↑

03 — Replay & diff

Confirm deterministic replay: re-run the exact inputs with the same seed, start compact, and expand only what changed.

Docs & runtime reference

Docs that match the demo. Open the Hello Fleet tutorial, core Concepts, CLI, and the embedded Runtime API—mapped 1:1 to what you just ran.

Show matching docs

Hello Fleet tutorial

Click-through instructions that mirror the demo run, receipts, and guardrail toggles.

Core concepts

Trust Mesh overview—North Star narrative, delivery/policy/replay pillars, and GTM guardrails.

CLI & receipts

fleetforge-ctl verify, capability tokens, and C2PA/SCITT manifests so CLI + UI stay aligned.

Roadmap & Status

Single execution tracker that ties roadmap themes to the Status & Acceptance evidence.

Why FleetForge vs alternatives?
  • Identity-scoped steps (capability tokens)
  • Signed receipts (C2PA + optional SCITT)
  • OPA→Wasm guardrails with replay budgets
Other adapters

This walkthrough uses LangGraph. AutoGen, CrewAI, and Semantic Kernel are supported—see the adapter matrix.