Manus billing importer (Butterfly Effect — reconciliation only)

What this is

The spendguard-importer-manus crate (D15) is a Rust binary + library that:

Pulls aggregate credit usage from the Manus admin REST API (/v1/usage) — feature-gated behind live so the default build is HTTP-free.
Converts vendor credits to estimated USD via the vendored TOML price table at services/importer_manus/assets/price_table.toml (integer micro-USD, saturating multiply).
Emits signed spendguard.audit.import.manus_credit CloudEvents tagged with reservation_source=subscription_meter (D14/D16 family-aligned) and import_source=manus_team_api — landing on the same dashboard surface as D13 subscription-meter rows and the sibling D14 Devin / D16 Genspark importers.

The default merge gate is a sanitized fixture replay — the committed tests/fixtures/manus_usage.json snapshot. Live mode is operator-opt-in.

Credit → $ conversion (three tiers)

| Tier | credit_cost_micro_usd | Audit row shape | |--------------------|-------------------------|-----------------| | team_plan | 20_526 (≈ $0.020526/credit) | $39/mo / 1900 credits, integer-truncated. Headline conversion: 47 credits × 20_526 = 964_722 micro-USD. | | enterprise | 0 (operator override required) | Negotiated per contract; dashboards show NULL spend until override lands. | | enterprise_byok | 0 (LOAD-BEARING) | BYOK tier customers pay the LLM provider direct. The importer MUST NOT double-bill them. |

The conversion is pure: credits × credit_cost_micro_usd in integer micro-USD, saturating at i64::MAX on overflow. The price table’s pricing_version is stamped on every emitted row at the moment of conversion — back-revisions to the rate file do not rewrite historical audit rows.

Why `enterprise_byok = 0` is load-bearing

BYOK customers bring their own LLM provider keys (Anthropic / OpenAI / etc.) and pay the LLM bill directly. If the importer charges them again on the Manus side, the SpendGuard dashboard double-counts the spend. The price table’s BYOK row is documented as load-bearing — never raise without re-reading the BYOK contract.

Install

Fixture mode (default merge gate)

cargo run -p spendguard-importer-manus --bin importer_manus -- \
    --mode fixture \
    --fixture services/importer_manus/tests/fixtures/manus_usage.json

The binary prints one CloudEvent per terminal session to stdout (in-progress sessions filtered by default — pass --include-in-progress to see them too). The demo runner script (deploy/demo/import_manus_fixture_demo.sh) wires the INSERT path against a throwaway demo postgres.

Live mode (operator-opt-in)

export MANUS_API_TOKEN=<Team+ admin API token>
export MANUS_API_BASE_URL=https://api.manus.ai   # optional override

cargo run -p spendguard-importer-manus --features live --bin importer_manus -- \
    --mode live

The live feature pulls reqwest with rustls-tls only (no native-tls, no openssl-sys), uses a 30s per-request timeout, and bounded exponential backoff (cap: 1 hour) for rate-limit resilience. Cursor pagination is hard-capped at 10_000 pages to prevent a misbehaving vendor next_cursor from looping forever.

Scope the token to the Manus Team+ admin API scope only — least privilege.

Idempotency

Re-running the same window does not double-emit. The CloudEvent event.id is a deterministic UUIDv5 derived from (workspace_id, session_id, window_end); the audit-row dedupe_key is manus:<session_id> (vendor-isolated namespace so D14 / D16 keys never collide); canonical_ingest dedups via the existing event_replay_dedup table.

You can re-run the demo as many times as you like:

make -C deploy/demo demo-import-manus-fixture
make -C deploy/demo demo-import-manus-fixture   # idempotent; 0 new rows

Sample CloudEvent

{
  "specversion": "1.0",
  "type": "spendguard.audit.import.manus_credit",
  "source": "spendguard-importer-manus",
  "id": "018f4b15-6d61-7c15-91fe-d0156d61aca0",
  "time": "2026-06-08T12:00:00Z",
  "datacontenttype": "application/json",
  "subject": "tenant/ws_FAKE_team_001/manus/workspace/ws_FAKE_team_001/session/ses_FAKE_team_completed_001",
  "data": {
    "schema_version": "v1alpha1",
    "tenant_id": "ws_FAKE_team_001",
    "workspace_id": "ws_FAKE_team_001",
    "session_id": "ses_FAKE_team_completed_001",
    "tier": "team_plan",
    "status": "completed",
    "credits_consumed": 47,
    "credit_cost_micro_usd": 20526,
    "amount_micro_usd": 964722,
    "pricing_version": "manus-credit-v1-2026-06",
    "model": "manus.session/credit",
    "input_tokens": 0,
    "output_tokens": 0,
    "window_start": "2026-06-05T14:00:00Z",
    "window_end": "2026-06-05T15:00:00Z",
    "reservation_source": "subscription_meter",
    "import_source": "manus_team_api",
    "ingestion_mode": "fixture",
    "fixture_provenance_sha256": "535d5cf688f486cac501436eb24e6357fa6c3677464001a0402e86686a50b0de",
    "dedupe_key": "manus:ses_FAKE_team_completed_001"
  }
}

Sample price table (TOML)

pricing_version = "manus-credit-v1-2026-06"
effective_from  = "2026-06-01T00:00:00Z"
currency        = "USD"

[tiers.team_plan]
credit_cost_micro_usd = 20526
# $39/mo / 1900 credits = $0.020526/credit; integer-truncated.

[tiers.enterprise]
credit_cost_micro_usd = 0
# Operator override required at deploy time.

[tiers.enterprise_byok]
credit_cost_micro_usd = 0
# LOAD-BEARING: BYOK customers pay LLM provider direct.

Why we cannot gate

The Manus agent loop runs entirely inside Butterfly Effect’s vendor-managed VM. The customer’s egress proxy never sees the LLM API call payload — the Manus VM dials OpenAI / Anthropic from inside the vendor’s own network boundary. The admin API exposes only post-hoc aggregate credit usage, not a pre-call hook.

This is Archetype IV in the SpendGuard framework-coverage analysis: a fully managed cloud agent where the customer has zero network visibility. The only feasible integration is reconciliation — pull the bill after the fact, surface it on the dashboard, alert on threshold crossings. Gating is architecturally impossible without Manus shipping a pre-call webhook.

What you see on the dashboard

Estimated $ per session, per workspace — labelled with pricing_version.
Credits consumed (always present, even for enterprise/BYOK zero-amount rows).
reservation_source = subscription_meter filter groups Manus spend alongside Devin (D14) and Genspark (D16) importer rows and Claude Code Pro / Codex (D13) advisory rows that skip the BYOK ledger.
Per-importer filter via import_source = manus_team_api.
Tier-level pivot: team_plan vs enterprise vs enterprise_byok.

Sibling importer integrations

D15 is part of a three-importer family that covers the agent-vendor Archetype IV ecosystem:

D14 Devin importer — Cognition Labs ACU usage.
D15 Manus importer (this page).
D16 Genspark importer — Genspark genspark_team_api usage.

All three follow the same locked invariants: reservation_source = subscription_meter, import_source = <vendor>_team_api, deterministic UUIDv5 event IDs, vendor-prefixed dedupe keys (devin: / manus: / genspark:), default build HTTP-free, live mode rustls-only behind the live feature flag.

Implementation pointers

Crate scaffold + Cargo.toml: services/importer_manus/Cargo.toml — workspace-excluded, publish = false, live feature gated.
Pure conversion: services/importer_manus/src/audit.rs (import_record_to_audit_row) — no I/O, no clock read, integer micro-USD only.
Price table: services/importer_manus/src/pricing.rs + assets/price_table.toml — saturating arithmetic, rejects negative credits.
CloudEvent builder: services/importer_manus/src/cloudevent_envelope.rs — golden-tested via tests/golden/cloudevent_v1alpha1_*.json.
Fixture loader: services/importer_manus/src/fixture.rs — hard-rejects non-synthetic IDs (ws_FAKE_* / ses_FAKE_*) at parse time.
Live client: services/importer_manus/src/live/client.rs — rustls-only reqwest wrapper with typed 401/403/429/5xx errors, bounded cursor pagination.
Migration: services/ledger/migrations/0060_import_source_widen_manus.sql — widens the existing audit_outbox.import_source CHECK to include manus_team_api.

For the full design rationale + slice-level breakdown, see docs/specs/coverage/D15_manus_importer/design.md.