Implement session-based API endpoints for creating, managing, and interacting with pre-funded sessions, including deposit and top-up invoice generation, macaroon authentication, and per-request debiting of compute costs.
Replit-Commit-Author: Agent
Replit-Commit-Session-Id: 418bf6f8-212b-4bb0-a7a5-8231a061da4e
Replit-Commit-Checkpoint-Type: full_checkpoint
Replit-Commit-Event-Id: 2dc3847e-7186-4a22-9c7e-16cd31bca8d9
Replit-Commit-Screenshot-Url: https://storage.googleapis.com/screenshot-production-us-central1/9f85e954-647c-46a5-90a7-396e495a805a/418bf6f8-212b-4bb0-a7a5-8231a061da4e/sPDHkg8
Replit-Helium-Checkpoint-Created: true
## New files
- btc-oracle.ts: CoinGecko BTC/USD fetch (60s cache), usdToSats() helper (ceil, min 1 sat),
5s abort timeout, fallback to BTC_PRICE_USD_FALLBACK env var (default $100k)
- lib/db/migrations/0002_cost_based_pricing.sql: SQL migration artifact adding 6 new columns
to jobs table (estimated_cost_usd, margin_pct, btc_price_usd, actual_input_tokens,
actual_output_tokens, actual_cost_usd); idempotent via ADD COLUMN IF NOT EXISTS
## Modified files
- pricing.ts: Full rewrite — per-model token rates (Haiku/Sonnet, env-var overridable),
DO infra amortisation per request, originator margin %, estimateInputTokens/Output by tier,
calculateActualCostUsd() for post-work ledger, async calculateWorkFeeSats() → WorkFeeBreakdown
- agent.ts: WorkResult now includes inputTokens + outputTokens from Anthropic usage;
workModel/evalModel exposed as readonly public; EVAL_MODEL/WORK_MODEL env var support
- lib/db/src/schema/jobs.ts: 6 new real/integer columns; schema pushed to DB
- jobs.ts route: Work invoice creation calls pricingService.calculateWorkFeeSats() async;
stores estimatedCostUsd/marginPct/btcPriceUsd; post-work stores actualInputTokens/
actualOutputTokens/actualCostUsd; GET response includes pricingBreakdown and costLedger
with totalTokens (input + output computed field)
- openapi.yaml: PricingBreakdown + CostLedger schemas (with totalTokens) added
- lib/api-zod/src/generated/api.ts: Regenerated with new schemas
- lib/api-client-react/src/generated/api.schemas.ts: Regenerated (PricingBreakdown, CostLedger)
- replit.md: 17 new env vars documented in cost-based pricing section
OpenAPI spec (lib/api-spec/openapi.yaml)
- Added POST /jobs, GET /jobs/{id}, GET /demo endpoints
- Added schemas: CreateJobRequest, CreateJobResponse, JobStatusResponse,
InvoiceInfo, JobState, DemoResponse, ErrorResponse
- Ran codegen: generated CreateJobBody, GetJobResponse, RunDemoQueryParams etc.
Jobs router (artifacts/api-server/src/routes/jobs.ts)
- POST /jobs: validates body, creates LNbits eval invoice, inserts job +
invoice in a DB transaction, returns { jobId, evalInvoice }
- GET /jobs/🆔 fetches job, calls advanceJob() helper, returns state-
appropriate payload (eval/work invoice, reason, result, errorMessage)
- advanceJob() state machine:
- awaiting_eval_payment: checks LNbits, atomically marks paid + advances
state via optimistic WHERE state='awaiting_eval_payment'; runs
AgentService.evaluateRequest, branches to awaiting_work_payment or rejected
- awaiting_work_payment: same pattern for work invoice, runs
AgentService.executeWork, advances to complete
- Any agent/LNbits error transitions job to failed
Demo router (artifacts/api-server/src/routes/demo.ts)
- GET /demo?request=...: in-memory rate limiter (5 req/hour per IP)
- Explicit guard for missing request param (coerce.string() workaround)
- Calls AgentService.executeWork directly, returns { result }
Dev router (artifacts/api-server/src/routes/dev.ts)
- POST /dev/stub/pay/:paymentHash: marks stub invoice paid in-memory
- Only mounted when NODE_ENV !== 'production'
Route index updated to mount all three routers
replit.md: documented full curl flow with all 6 steps, demo endpoint,
and dev stub-pay trigger
End-to-end verified with curl:
- Full flow: create → eval pay → evaluating → work pay → executing → complete
- Error cases: 400 on missing body/param, 404 on unknown job
