Skip to content

CHANGELOG.md

Latest commit

 

History

History
573 lines (522 loc) · 51.9 KB

CHANGELOG.md

File metadata and controls

573 lines (522 loc) · 51.9 KB

Changelog

All notable changes to Maschina will be documented here.

Format: Semantic Versioning — [version] YYYY-MM-DD

[Unreleased]

Added (2026-03-15 — Comprehensive test suite / feat/testing)

  • API unit tests — services/api/src/routes/health.test.ts, auth.test.ts, agents.test.ts covering Layer 1 (401 auth rejection) and Layer 2 (400 validation errors) without DB/network
  • Plans unit tests — packages/plans/src/gates.test.ts — isAtLeastTier, nextTier, can.skipBilling, can.useApiKeys, can.useCompliance, can.inviteTeamMembers, hasFeature
  • Validation unit tests — packages/validation/src/schemas.test.ts — assertValid, parseBody, ValidationError, all agent + auth + run schemas
  • Model access integration tests — packages/model/src/model_access.integration.test.ts — plan tier × model access matrix, resolveModel cascade fallback for all tiers
  • Rust scheduler unit tests — 14 tests in services/daemon/src/scheduler/mod.rs #[cfg(test)] module: score() (capacity, load, model match, GPU, stake cap), weighted_random_pick() (empty, single, zero-total, probability distribution)
  • Python risk tests — packages/risk/tests/test_risk.py replaces placeholder: check_input (injection patterns, length limit), check_output (PII detection), check_quota_pre_run (exhausted, low, unlimited)
  • Python agent tests — packages/agents/tests/test_agents.py replaces placeholder: AgentType enum, all 5 concrete agents, ABC enforcement, to_dict shape
  • Python runtime model tests — packages/runtime/tests/test_runtime.py replaces placeholder: RunInput, RunResult, Message, ToolResult validation and serialization
  • Integration test suite (tests/integration/) — auth_flow.test.ts (full createApp() middleware stack), security.test.ts (auth bypass, injection payloads, security headers), chaos.test.ts (DB fault, Redis fault, concurrent request handling)
  • E2E smoke tests (tests/e2e/smoke.spec.ts) — skipped unless RUN_E2E=true
  • k6 load test scripts (tests/load/) — health.k6.js (ramp 20→100 VUs, p95<100ms), auth.k6.js (constant-arrival + spike, rate limit detection), agents.k6.js (CRUD lifecycle)
  • tests/vitest.config.ts — vitest config for integration/e2e suites
  • tests/tsconfig.json — TypeScript config for tests package

Changed (2026-03-15 — Comprehensive test suite / feat/testing)

  • tests/package.json — updated scripts to use vitest config, added hono devDependency
  • Deleted old placeholder test files (test_agents_placeholder.py, test_risk_placeholder.py, test_runtime_placeholder.py) — replaced with real implementations

Added (2026-03-15 — Chain wiring / feat/chain-wiring)

  • services/daemon/src/chain.rs — on-chain receipt anchoring: computes deterministic SHA-256 payload hash, builds Anchor anchor_receipt instruction (discriminator + borsh args), derives receipt + pool PDAs, loads authority keypair, submits via Helius RPC; fire-and-forget
  • services/daemon/src/orchestrator/analyze.rs — calls chain::submit_receipt() after every successful run; converts task price cents → USDC lamports; no-op when CHAIN_ENABLED=false
  • services/daemon/src/config.rs — added helius_rpc_url, settlement_program_id, solana_authority_keypair, chain_enabled fields
  • services/daemon/src/main.rs — registered mod chain
  • services/daemon/Cargo.toml — added solana-sdk, solana-rpc-client, borsh

Added (2026-03-15 — Auto-suspend / feat/auto-suspend)

  • services/api/src/routes/nodes.ts — slash handler now auto-suspends node (status = 'suspended', suspendedAt) when post-slash stakedUsdc drops below the tier minimum; response includes suspended: true flag
  • services/daemon/src/orchestrator/analyze.rs — update_node_reputation() now auto-suspends active nodes whose reputation_score drops below 20 with ≥10 tasks of signal; prevents degraded nodes from continuing to receive work

Added (2026-03-15 — Solana settlement program / feat/solana-program)

  • programs/settlement/src/instructions/mod.rs — exports all 5 instruction modules
  • programs/settlement/src/instructions/anchor_receipt.rs — AnchorReceipt context + handler: Ed25519 signature verification (stub), receipt PDA init, earnings split (65/20/10/5) accumulated into SettlementPool, ReceiptAnchored event emitted
  • programs/settlement/src/instructions/deposit_stake.rs — DepositStake: NodeStake + SettlementPool PDAs init_if_needed on first deposit; staked_amount accumulated; StakeDeposited event
  • programs/settlement/src/instructions/withdraw_stake.rs — WithdrawStake (7-day lock) + FinaliseWithdrawal (post-lock SPL transfer stub); Clock::get() validation
  • programs/settlement/src/instructions/slash_stake.rs — SlashStake: bps validation (1–10000), proportional slash from staked_amount, total_slashed tracking; StakeSlashed event
  • programs/settlement/src/instructions/settle_earnings.rs — SettleEarnings: drains pool to node/developer/treasury/validators wallets (SPL CPI stub); EarningsSettled event
  • programs/settlement/src/lib.rs — added finalise_withdrawal instruction
  • programs/Anchor.toml — settlement program registered under [programs.localnet]
  • packages/chain/src/settlement.ts — TypeScript Anchor client: PDA helpers (receiptPda, stakePda, poolPda), getSettlementProgram, fetchReceipt, isReceiptAnchored, UUID<->bytes utilities
  • packages/chain/src/webhook.ts — Helius webhook processing: HeliusWebhookPayload types, processHeliusWebhook dispatcher, registerSettlementWebhook helper
  • packages/chain/src/index.ts — exports for settlement + webhook modules
  • packages/chain/package.json — added @coral-xyz/anchor ^0.30.1
  • services/api/src/routes/webhooks.ts — POST /webhooks/helius: Helius auth header verification, processHeliusWebhook dispatch, onReceiptAnchored handler stub
  • services/api/src/env.ts — HELIUS_WEBHOOK_SECRET optional env var

Added (2026-03-15 — Run streaming / feat/run-streaming)

  • services/runtime/src/streaming.py — SSE streaming endpoint for all three runners (Anthropic, OpenAI, Ollama); chunk/done/error event types; POST /stream added to FastAPI app
  • services/realtime/src/handlers.rs — POST /internal/run-event handler: receives run status + chunk events from daemon and fans out to connected clients via broadcast registry
  • services/realtime/src/main.rs — /internal/run-event route registered

Added (2026-03-15 — Solana foundation / feat/solana-foundation)

  • packages/chain/ — Helius + Solana foundation: client singleton, wallet validation, Ed25519 ownership verification (buildChallenge + verifyWalletSignature via nacl)
  • services/api/src/routes/wallets.ts — GET /wallets/challenge, POST /wallets, GET /wallets, POST /wallets/verify, DELETE /wallets/:id; Helius + Orb chosen as Solana RPC + usage metering layers
  • CI + api package.json updated for @maschina/chain

Added (2026-03-15 — Node binary / feat/node-binary)

  • services/node/ — new maschina-node Rust binary for compute node operators:
    • src/identity.rs — Ed25519 keypair generation + persistence to ~/.config/maschina-node/identity.toml; sign() stub for Phase 5 receipt signing
    • src/config.rs — env-driven config: MASCHINA_API_URL, MASCHINA_API_KEY, NODE_NAME, NODE_REGION, NODE_INTERNAL_URL, NODE_HEARTBEAT_INTERVAL_SECS, NODE_MAX_CONCURRENT_TASKS, NODE_CONFIG_DIR
    • src/api.rs — typed API client: register_node(), submit_public_key(), heartbeat()
    • src/heartbeat.rs — periodic heartbeat loop with ActiveTaskCounter + send_once() for startup/shutdown
    • src/main.rs — full lifecycle: load/generate identity → register (idempotent) → submit public key → initial heartbeat → heartbeat loop → graceful shutdown
  • Cargo.toml (workspace) — services/node added to workspace members
  • .github/workflows/ci.yml — maschina-node added to all 4 Rust CI steps (fmt, clippy, test, build)

Added (2026-03-15 — Task watchdog / feat/task-watchdog)

  • services/daemon/src/watchdog.rs — standalone watchdog loop that sweeps agent_runs every 30s for runs stuck in running status beyond the timeout threshold; force-fails them with error_code = "watchdog_timeout"; reuses analyze phase for reputation update
    • realtime notification; race-safe (UPDATE WHERE status = 'running' + rows_affected check)
  • services/daemon/src/config.rs — watchdog_timeout_secs: Option<i64> field; configurable via WATCHDOG_TIMEOUT_SECS env var; defaults to 600s
  • services/daemon/src/main.rs — watchdog spawned alongside orchestrator + health server; mod watchdog registered

Added (2026-03-15 — Node identity + staking / feat/node-identity)

  • packages/db/src/schema/pg/enums.ts — stakeEventTypeEnum ("deposit" | "withdraw" | "slash")
  • packages/db/src/schema/pg/nodes.ts — publicKey: text column on nodes table (Ed25519 public key hex, 64 chars); nodeStakeEvents table: append-only staking ledger with amountUsdc, balanceAfterUsdc, reason, triggeredBy, slashPct, txSignature columns; NodeStakeEvent / NewNodeStakeEvent type exports
  • packages/db/src/schema/pg/receipts.ts — nodeSignature: text (Ed25519 sig, nullable) and signingAlg: text (null until node binary ships; "ed25519" when present) columns on execution_receipts — backward-compatible alongside existing HMAC signature column
  • services/api/src/routes/nodes.ts — 5 new endpoints:
    • POST /nodes/:id/public-key — node submits Ed25519 public key (idempotent, supports rotation)
    • POST /nodes/:id/stake — record USDC deposit, updates nodes.stakedUsdc atomically
    • POST /nodes/:id/unstake — stake withdrawal; validates balance stays >= tier minimum
    • POST /nodes/:id/slash — admin-triggered slash (slashPct%); inserts stake event
    • GET /nodes/:id/stake — returns balance + stake event history (paginated) Stake minimums enforced: micro=0, edge=100, standard=500, verified=5000, datacenter=25000 USDC

Added (2026-03-15 — Node earnings / feat/node-earnings)

  • packages/db/src/schema/pg/nodes.ts — nodeEarnings table: append-only per-run earnings ledger with 65/20/10/5 split columns (nodeCents, developerCents, treasuryCents, validatorCents), billing multiplier, token counts, settlement status; indexed by node + status
  • services/daemon/src/orchestrator/analyze.rs — record_node_earnings() fires after every successful run; billing_multiplier() lookup by model prefix; task_price_cents() = tokens/1k × 0.2¢ × multiplier + 1¢/run; splits total into 65/20/10/5 and inserts into node_earnings (fire-and-forget)
  • services/api/src/routes/nodes.ts — GET /nodes/:id/earnings: returns per-run earnings rows + totalPendingCents + totalSettledCents; node owner or admin only; filterable by status, paginated

Added (2026-03-15 — Scheduler v2 + revenue split / feat/scheduler-v2)

  • services/daemon/src/scheduler/mod.rs — reputation and stake factored into node scoring: (reputation_score / 100) * 20 pts + min(stake / 1000, 1) * 5 pts on top of existing load (50) + model match (30) + GPU (20) factors; max score ~125
  • packages/marketplace/src/index.ts — added calcExecutionRevenue() with the correct 65/20/10/5 split (node/developer/treasury/validators) for per-execution on-chain task revenue; calcRevenueShare() retained for fiat listing sales (70/30)

Added (2026-03-14 — Marketplace payments / feat/marketplace-payments)

  • packages/billing/src/marketplace.ts — createMarketplacePaymentIntent(): creates a pending order + Stripe PaymentIntent for a paid listing; returns {clientSecret, paymentIntentId, orderId} for Stripe.js confirmation
  • packages/billing/src/webhooks.ts — handles payment_intent.succeeded: detects maschinaProduct: "marketplace_listing", completes the order, credits seller 70% via credit ledger, increments download count, forks the agent config as a new agent owned by the buyer (idempotent)
  • packages/billing/src/index.ts — exports new marketplace helpers
  • services/api/src/routes/marketplace.ts:
    • POST /marketplace/listings/:id/buy — creates PI, returns clientSecret
    • GET /marketplace/orders — buyer purchase history with listing names
    • GET /marketplace/earnings — seller earnings total + transaction list

Added (2026-03-15 — Workflow wiring / feat/workflow-wiring)

  • services/worker/src/worker/workflows/activities.py — run_agent_step fully wired:
    • Fetches agent config (system_prompt, model) + enabled skills from DB via asyncpg
    • Fetches user plan_tier from subscriptions table
    • Generates per-step run_id (UUID) for runtime tracing
    • Fixes payload to match RunRequest: input_payload: {"message": prompt} (was "prompt")
    • Includes all required fields: run_id, plan_tier, model, system_prompt, max_tokens, timeout_secs, skills, skill_configs
    • Returns normalised {"step_id", "step_run_id", "output", "output_payload", "input_tokens", "output_tokens"}
  • packages/sdk/ts/tsconfig.json — added "examples" to include array (fixes red files in IDE)

Added (2026-03-15 — Agent collaboration / feat/agent-collaboration)

  • packages/runtime/src/maschina_runtime/tools.py — DelegateAgentTool: delegates a subtask to another agent via POST /internal/delegate, returns output synchronously; guards against self-delegation
  • packages/runtime/src/maschina_runtime/__init__.py — exports DelegateAgentTool
  • services/runtime/src/config.py — MASCHINA_API_URL + INTERNAL_SECRET settings
  • services/runtime/src/skills.py — delegate_agent slug wired; build_tools() gains caller_agent_id + user_id params for delegation context
  • services/runtime/src/runner.py — passes caller_agent_id + user_id to build_tools()
  • services/api/src/routes/internal.ts — POST /internal/delegate: secret-gated, fetches target agent config + skills, calls runtime synchronously, returns output
  • services/api/src/routes/agents.ts — GET /agents/discover: lists own agents available for delegation (filterable by type)
  • services/api/src/app.ts — /internal routes registered
  • packages/connectors/src/skills.ts — delegate_agent added to SKILL_CATALOG (access+ tier)
  • services/runtime/tests/test_runner_routing.py — DelegateAgentTool added to maschina_runtime.tools stub

Added (2026-03-15 — Reputation / feat/reputation)

  • services/daemon/src/scheduler/mod.rs — select_node() now returns (String, Option<Uuid>) (url + node_id)
  • services/daemon/src/runtime/mod.rs — dispatch() renamed to dispatch_to(state, run, node_url) — takes explicit URL
  • services/daemon/src/orchestrator/execute.rs — calls scheduler separately; passes node_id to finalize_run
  • services/daemon/src/orchestrator/analyze.rs — finalize_run takes node_id: Option<Uuid>; fires update_node_reputation + update_agent_reputation (tokio::spawn, fire-and-forget) after every run outcome
  • Node reputation formula: completed / (completed + failed + timed_out) * 100, clamped 0–100, frozen at current score until 5+ total tasks
  • packages/db/src/schema/pg/agents.ts — added totalRunsCompleted, totalRunsFailed, reputationScore columns
  • packages/db/src/schema/sqlite/agents.ts — same columns mirrored for local dev
  • packages/marketplace/src/index.ts — ListingDoc + listingToDoc() gains optional reputationScore field
  • packages/search/src/indexes.ts — marketplace index: reputationScore added to filterable + sortable + displayed attributes
  • services/api/src/routes/marketplace.ts — publish endpoint fetches agent reputationScore and includes it in Meilisearch upsert

[Unreleased - cascade-fallback]

Added (2026-03-15 — Analytics / feat/analytics)

  • packages/analytics/src/posthog.ts — PostHog client, lazy-init, typed event helpers (agent.created, agent.run.completed, connector.installed, subscription.upgraded, etc.)\n- packages/analytics/src/langsmith.ts — LangSmith HTTP tracing client (startTrace, endTrace, failTrace)\n- packages/analytics/src/index.ts — package exports\n- packages/analytics/package.json + tsconfig.json — new package\n- services/api/src/routes/analytics.ts — GET /analytics/overview, /runs, /tokens, /agents/top (M5+ gate)\n- services/api/src/app.ts — analytics routes registered\n- services/runtime/src/tracing.py — LangSmith async trace wrapper\n- services/runtime/src/runner.py — start/end/fail trace around every agent run\n\n### Added (2026-03-15 — Connector integrations / feat/connectors)\n- packages/connectors/src/definitions.ts — connector catalog: Slack (OAuth), GitHub (OAuth), Notion (API key), Linear (API key)\n- packages/connectors/src/crypto.ts — AES-256-GCM encrypt/decrypt for credential storage\n- packages/connectors/src/skills.ts — added slack, github, notion, linear to SKILL_CATALOG\n- packages/connectors/src/index.ts — exports crypto + definitions\n- services/api/src/routes/connectors.ts — full connector API: definitions list, CRUD, OAuth flow, incoming webhook receivers (Slack, GitHub, Linear signature verification)\n- services/api/src/app.ts — connector routes registered at /connectors\n- packages/runtime/src/maschina_runtime/tools.py — SlackTool, GitHubTool, NotionTool, LinearTool\n- services/runtime/src/skills.py — slack, github, notion, linear wired into build_tools()

Added (2026-03-14 — Organization management / feat/orgs)

  • packages/validation/src/schemas/org.ts — CreateOrgSchema, UpdateOrgSchema, InviteMemberSchema, UpdateMemberRoleSchema
  • packages/db/src/schema/pg/relations.ts — organizationsRelations, organizationMembersRelations, organizationInvitesRelations
  • services/api/src/routes/orgs.ts — full org API: POST /orgs, GET /orgs, GET /orgs/:id, PATCH /orgs/:id, DELETE /orgs/:id, GET /orgs/:id/members, PATCH /orgs/:id/members/:memberId, DELETE /orgs/:id/members/:memberId, POST /orgs/:id/invites, GET /orgs/:id/invites, DELETE /orgs/:id/invites/:inviteId, POST /orgs/invites/:token/accept, GET /orgs/:id/agents, GET /orgs/:id/usage
  • services/api/src/app.ts — org routes registered at /orgs

Added (2026-03-14 — Agent sandboxing)

  • packages/runtime/src/maschina_runtime/tools.py — CodeExecTool gains memory_limit_mb + cpu_limit_secs; applies resource.setrlimit (RLIMIT_AS, RLIMIT_CPU, RLIMIT_FSIZE) via preexec_fn on Unix; no-op on Windows
  • services/runtime/src/config.py — SANDBOX_ENABLED, SANDBOX_MEMORY_LIMIT_MB, SANDBOX_CPU_LIMIT_SECS settings
  • services/runtime/src/skills.py — passes sandbox limits from config when constructing CodeExecTool
  • services/runtime/src/models.py — RunResponse gains sandbox_type: str | None ("subprocess_rlimit" | "subprocess" | None)
  • services/runtime/src/runner.py — sets sandbox_type in response based on active skills + platform
  • services/daemon/src/runtime/mod.rs — RunOutput gains sandbox_type: Option<String>
  • services/daemon/src/orchestrator/analyze.rs — persist_success writes sandbox_type to agent_runs table
  • .env.example — sandbox env vars documented

Added (2026-03-14 — Agent permissions)

  • packages/db/src/schema/pg/enums.ts — agentPermissionEnum: internet_access, code_execution, external_api, file_read, file_write, memory_read, memory_write, send_email, send_webhook
  • packages/db/src/schema/pg/agents.ts — agentPermissions table (agent_id, permission, granted_at, granted_by_user_id); unique index on (agent_id, permission)
  • packages/db/src/schema/pg/relations.ts — agentPermissionsRelations; agentsRelations gains permissions many-relation
  • services/daemon/src/error.rs — PermissionDenied variant
  • services/daemon/src/orchestrator/evaluate.rs — check_skill_permissions(): code_exec → code_execution, web_search/http_fetch → internet_access; blocks run if permission absent
  • services/api/src/routes/permissions.ts — GET/PUT /agents/:id/permissions, DELETE /agents/:id/permissions/:permission
  • services/api/src/app.ts — permission routes registered

Added (2026-03-14 — Proof of Compute verification)

  • services/api/src/routes/receipts.ts — POST /receipts/:id/verify: re-derives HMAC-SHA256 from stored payload and confirms signature matches; returns { valid: boolean, receiptId }
  • .env.example — PROOF_SECRET documented with generation hint

Added (2026-03-14 — Proof of Compute)

  • packages/db/src/schema/pg/receipts.ts — execution_receipts table (run_id, agent_id, user_id, node_id, model, input_tokens, output_tokens, payload, signature, issued_at)
  • packages/db/src/schema/pg/relations.ts — executionReceiptsRelations; agentRunsRelations gains receipt many-relation
  • services/daemon/src/receipt.rs — HMAC-SHA256 signing: canonical JSON payload (sorted keys) → hex signature; issue_receipt() inserts receipt post-run (non-fatal)
  • services/daemon/src/orchestrator/analyze.rs — issue_receipt() called after persist_success on every completed run
  • services/daemon/src/config.rs — proof_secret field (env: PROOF_SECRET, dev fallback)
  • services/daemon/Cargo.toml — hmac, sha2, hex deps
  • services/api/src/routes/receipts.ts — GET /receipts/:id, GET /agents/:agentId/receipts; ownership-gated per userId
  • services/api/src/app.ts — receipt routes registered

Added (2026-03-14 — Feature flags)

  • packages/flags/src/flags.ts — flag registry (marketplaceEnabled, workflowsEnabled, memoryEnabled, proofOfComputeEnabled, nodeRegistrationEnabled, distributedComputeEnabled, machTeamPlanVisible, billingEnabled, skillMarketplaceEnabled, pluginsEnabled, newRunUiEnabled); FlagName union type
  • packages/flags/src/types.ts — FlagContext (userId, orgId, tier, email, attributes), FlagValue, FlagKey
  • packages/flags/src/client.ts — FlagClient.is() / FlagClient.all(); getFlags(ctx) — LaunchDarkly (lazy) → Redis cache (TTL 60s) → defaults; isEnabled() convenience helper
  • packages/flags/src/flags.test.ts — tests: defaults, overrides, all(), flag shape
  • .github/workflows/ci.yml — @maschina/storage and @maschina/flags added to TS build chain in both ts-typecheck and ts-test jobs

Added (2026-03-14 — S3/CloudFront object storage)

  • packages/storage/src/client.ts — StorageClient: upload, uploadJson, download, downloadJson, delete, presignedDownload, presignedUpload, publicUrl; CloudFront URL rewriting; MinIO-compatible via S3_ENDPOINT; singleton getStorage()
  • packages/storage/src/keys.ts — StorageKeys: agentArtifact, taskOutput, upload path helpers
  • packages/storage/src/storage.test.ts — 12 tests (StorageKeys + publicUrl)
  • packages/storage/package.json — @aws-sdk/client-s3 + @aws-sdk/s3-request-presigner
  • services/api/src/routes/storage.ts — POST /storage/upload-url, GET /storage/download-url, DELETE /storage/object; ownership-gated per userId
  • services/api/src/app.ts — storage routes registered
  • services/api/Dockerfile — storage package built in image

Added (2026-03-14 — Multi-agent workflows)

  • packages/db/src/schema/pg/workflows.ts — workflows + workflowRuns tables; workflowTypeEnum + workflowRunStatusEnum added to enums
  • packages/events/src/types.ts — WorkflowRunQueued + WorkflowRunCancelled subjects + data types
  • services/worker/src/worker/workflows/ — Temporal workflow + activities:
    • agent_workflow.py — AgentWorkflow (@workflow.defn): sequential, parallel, conditional execution strategies
    • activities.py — run_agent_step (calls runtime /run), update_run_status (writes to DB)
    • temporal_worker.py — Temporal worker runner (task queue: maschina-workflows)
  • services/worker/src/worker/handlers/workflow.py — NATS handler that starts Temporal workflow via client
  • services/worker/src/worker/main.py — asyncio.gather(run_consumer(), run_temporal_worker())
  • services/worker/pyproject.toml — added temporalio>=1.7
  • services/worker/src/worker/config.py — added temporal_url + runtime_url
  • services/api/src/routes/workflows.ts — full workflow API (CRUD + trigger + run status + cancel)
  • services/api/src/app.ts — registered workflow routes at /workflows

Added (2026-03-14 — Agent marketplace)

  • packages/marketplace/src/index.ts — calcRevenueShare() (70/30 split), listingToDoc() (Meilisearch shape), generateSlug()
  • packages/db/src/schema/pg/marketplace.ts — agentId (nullable FK) + agentConfig (jsonb snapshot) on marketplace_listings; full marketplaceOrders + marketplaceReviews tables
  • services/api/src/routes/marketplace.ts — full marketplace API: browse (Meilisearch + DB fallback), get listing+reviews, create/update/publish/unpublish listing, fork agent (copies config snapshot), submit review (requires completed order), seller's own listings
  • services/api/src/app.ts — registered marketplace routes at /marketplace
  • services/api/package.json — added @maschina/marketplace workspace dep

Added (2026-03-14 — Resource scheduler)

  • services/daemon/src/scheduler/mod.rs — scored node selection replacing naive "most recent heartbeat" strategy
    • Queries all active nodes with fresh heartbeat in one JOIN (nodes + node_capabilities + latest heartbeat via LATERAL)
    • Scoring: load factor (50pts, proportional to spare capacity), model match (30pts, prefix match on supported_models), GPU bonus (20pts, for ollama/* models on GPU nodes)
    • Nodes at full capacity (active_task_count >= max_concurrent_tasks) are excluded from selection
    • Falls back to RUNTIME_URL when no nodes available or all at capacity; logs capacity state at DEBUG
  • services/daemon/src/runtime/mod.rs — replaced select_node_url() with crate::scheduler::select_node(state, model); removed dead NodeRow struct
  • services/daemon/src/main.rs — registered scheduler module

Added (2026-03-14 — Agent skill framework)

  • packages/db/src/schema/pg/agents.ts — agent_skills table (agent_id, skill_name, config, enabled)
  • packages/connectors/src/skills.ts — skill catalog: http_fetch (access+), web_search (m1+, Brave API), code_exec (m5+, sandboxed subprocess); tier gating helpers canUseSkill(), listSkills()
  • packages/connectors/src/index.ts + package.json — wired up, deps fixed to @maschina/plans
  • packages/runtime/src/maschina_runtime/tools.py — WebSearchTool (Brave Search API, max_results configurable), CodeExecTool (subprocess sandbox, timeout configurable, 30s max); HttpFetchTool gains optional domain allowlist
  • services/runtime/src/skills.py — maps skill slugs → Tool instances; reads BRAVE_SEARCH_API_KEY from env
  • services/runtime/src/models.py — RunRequest gains skills: list[str] + skill_configs: dict
  • services/runtime/src/runner.py — builds tools from req.skills and passes to AgentRunner (Anthropic only)
  • services/daemon/src/orchestrator/scan_compat.rs — JobToRun gains skills: Vec<String> + skill_configs
  • services/daemon/src/orchestrator/scan.rs — initializes empty skills on JobToRun
  • services/daemon/src/orchestrator/evaluate.rs — queries agent_skills table, populates run.skills + run.skill_configs before execute
  • services/daemon/src/runtime/mod.rs — RuntimeRequest gains skills + skill_configs, passed through to Python runtime
  • services/api/src/routes/skills.ts — GET /skills (catalog), GET /agents/:id/skills, PUT /agents/:id/skills/:slug, DELETE /agents/:id/skills/:slug; tier-gated on upsert
  • services/api/src/app.ts — skill routes registered
  • .env.example — BRAVE_SEARCH_API_KEY added

Added (2026-03-14 — Agent memory)

  • packages/vector/src/collections.ts — added agent_memory collection (1536-dim Cosine)
  • services/runtime/src/memory.py — episodic memory: retrieve top-k similar memories (Qdrant + OpenAI text-embedding-3-small), store output after each run; all errors swallowed gracefully
  • services/runtime/src/runner.py — retrieve memories before routing to LLM, inject as memory block in system prompt; store output memory after run
  • services/runtime/src/config.py — added qdrant_url, qdrant_api_key, memory_enabled, memory_top_k settings
  • services/runtime/pyproject.toml — added qdrant-client>=1.9 dependency
  • services/api/src/routes/memory.ts — GET /agents/:id/memory (scroll with pagination), DELETE /agents/:id/memory (clear all); ownership-gated, graceful on Qdrant unreachable
  • services/api/src/app.ts — memory routes registered under /agents
  • apps/docs/api-reference/agents.mdx — Agent Memory section added (list + clear endpoints)

Added (2026-03-14 — Node registry API)

  • services/api/src/routes/nodes.ts — full node management CRUD: POST /nodes/register, POST /nodes/:id/heartbeat, GET /nodes, GET /nodes/:id, PATCH /nodes/:id, DELETE /nodes/:id
  • services/daemon/src/runtime/mod.rs — select_node_url() queries nodes table for most-recently-healthy active node (heartbeat <60s), falls back to RUNTIME_URL on miss or error
  • services/api/src/app.ts — /nodes route registered

Added (2026-03-14 — Meilisearch search)

  • packages/search/src/indexes.ts — agents index settings updated: type and status added to searchable/filterable/displayedAttributes
  • packages/search/src/search.test.ts — unit tests for INDEXES structure, module exports, client singleton
  • services/api/src/routes/search.ts — GET /search route: auth-scoped, supports q, type, limit, offset; graceful Meilisearch degradation (returns empty result on unreachable)
  • services/api/src/routes/agents.ts — Meilisearch sync on agent create/update/delete (fire-and-forget, non-blocking)
  • services/api/src/app.ts — /search route registered
  • services/api/src/index.ts — ensureIndexes() called at startup (non-fatal on failure)
  • services/api/package.json — @maschina/search added as dependency

Added (2026-03-08 — Model routing)

  • packages/model/src/catalog.ts — TypeScript model catalog: 3 Anthropic cloud models + 3 Ollama local models, per-tier access gates, billing multipliers (Haiku 1x, Sonnet 3x, Opus 15x, Ollama 0x)
  • packages/model/src/index.ts — Barrel export
  • packages/model/src/catalog.test.ts — 20 vitest tests covering multipliers, tier access, validation, resolution
  • packages/model/tsconfig.json + build script — TS package alongside existing Python code
  • packages/validation — RunAgentSchema gains optional model field
  • packages/jobs — AgentExecuteJob gains model + systemPrompt fields; dispatchAgentRun updated
  • services/api — Model access validation at run dispatch; resolves system prompt from agent.config.systemPrompt; passes model + system prompt through job queue
  • services/daemon — AgentExecuteJob, JobToRun gain model + system_prompt; RuntimeRequest now sends all fields the Python runtime needs (plan_tier, model, system_prompt, max_tokens, timeout_secs); URL fixed from /execute → /run
  • services/daemon — RunOutput.payload renamed to output_payload to match Python RunResponse
  • services/runtime — Full model routing in runner.py: routes by model ID prefix (ollama/* vs Anthropic), applies billing multiplier, lazy-imports Anthropic client per request; drops global Ollama flag
  • services/runtime/tests/test_runner_routing.py — Unit tests for multiplier + routing helpers (no real LLM calls)
  • CI + pytest scripts updated to include services/runtime tests

Fixed (2026-03-08 — Model routing)

  • Daemon was calling /execute endpoint on Python runtime — correct endpoint is /run
  • Daemon RuntimeRequest was missing plan_tier, model, system_prompt, timeout_secs fields that the Python RunRequest model requires

Fixed (2026-03-07 — Session N+1: backend boot + E2E)

  • All 31 TS packages now build clean (pnpm turbo build --filter='./packages/*')
  • packages/cache/src/client.ts — ioredis ESM default import via (Redis as any) constructor cast
  • packages/cache/src/ops.ts — import type { Redis as RedisType } (named type import)
  • packages/validation/src/schemas/agent.ts — agent type enum updated to match DB: signal/analysis/execution/optimization/reporting
  • services/api/src/env.ts — dotenv loaded via config() at module load; Stripe keys optional with empty-string default
  • services/api/src/routes/auth.ts — validatePasswordStrength returns PasswordValidation object; check .valid not truthiness; verifyPassword args were swapped (fixed to verifyPassword(hash, plain))
  • services/api/src/middleware/auth.ts — resolveAuth accepts Headers object; was incorrectly passing raw string; fixed to c.req.raw.headers
  • services/api/src/routes/billing.ts — /billing/portal now returns 400 for internal/access tier users (no Stripe customer)
  • All API route files — changed from "drizzle-orm" → from "@maschina/db" for consistent single-instance resolution
  • packages/telemetry/src/sdk.ts — PeriodicExportingMetricReader moved to @opentelemetry/sdk-metrics
  • packages/nats/src/streams.ts — readonly string[] spread to mutable array
  • packages/plans/src/gates.ts — ?? null for PlanTier | null return

Added (2026-03-07 — TUI launcher)

  • packages/cli/src/tui.rs — ratatui TUI launcher (maschina with no args):
    • Horizontally centered panel (64% width) with rounded borders
    • Service status panel: api (:3000), gateway (:8080), realtime (:4000), runtime (:8000), daemon
    • Status detection: PID file check → port check → stopped
    • PID files at ~/.local/share/maschina/pids/
    • Auto-refresh every 3s; message banner expires after 5s
    • Keys: ↑↓/jk navigate, s start, x stop, a start all, X stop all, r refresh, q/Esc quit
    • Workspace detection: start/stop only available when in Maschina project root
    • Log files written to .maschina/logs/<service>.log
    • Monochrome palette: White/Gray/DarkGray, no color theme
  • packages/cli/src/main.rs — command is now Option<Commands>; None launches TUI

Added (2026-03-07 — CLI + install script)

  • packages/cli — complete rewrite of maschina CLI:
    • maschina setup — interactive wizard: API URL, email/password or API key, credential validation, project .maschina/ init
    • maschina login / maschina logout — auth commands
    • maschina status — connection + account info
    • maschina doctor — config, connectivity, and project health checks
    • maschina agent list|deploy|stop|run — agent management with correct API types
    • maschina keys list|create|revoke — API key management
    • maschina usage — quota usage with visual bar chart
    • maschina logs <run_id> — run inspection
    • --json global flag for scripting/CI output
    • --profile <name> for multi-environment configs (each profile: ~/.config/maschina/<profile>.toml)
    • arg_required_else_help = true (shows help instead of panicking on no args)
    • src/output.rs — unified human/JSON output layer
    • src/project.rs — .maschina/config.toml per-project config (name, description, agent defaults, runtime URL)
  • install.sh — curl installer:
    • Detects OS (macOS/Linux) and architecture (x86_64/aarch64)
    • Checks required dependencies (curl, tar)
    • Downloads release binary from GitHub
    • Installs to /usr/local/bin or ~/.local/bin
    • Adds to PATH in shell rc file (zsh/bash/fish)
    • Runs maschina setup automatically if interactive TTY

Added (2026-03-07 — CLI full build)

  • packages/cli/src/services.rs — shared service management module (PID files, port checks, probe, start/stop, log paths); start_svc checks ~/.local/share/maschina/bin/ first (installed), falls back to workspace dev mode
  • packages/cli/src/tui.rs — complete rewrite: borderless two-zone launcher (Services + Menu), deimos-style centered Rect, ▸ cursor, spinner animation, responsive column collapse, Tab/j/k/s/x/l/Enter/r/q controls; returns LaunchTarget for post-TUI dispatch
  • packages/cli/src/commands/service.rs — start/stop/restart/status/logs using shared services:: module; --json support
  • packages/cli/src/commands/setup.rs — 5-step interactive wizard: connection → account (login/register/paste) → AI providers (MultiSelect: Anthropic/OpenAI/Ollama/OpenRouter/Gemini/Mistral) → database (SQLite/Postgres/Neon) → workspace init
  • packages/cli/src/commands/agent.rs — added runs() (run history table) and inspect() (detailed agent view)
  • packages/cli/src/config.rs — added db_url: Option<String> and model_providers: Vec<ModelProvider>
  • packages/cli/src/main.rs — full rewrite: comprehensive command tree (setup/login/logout/status/doctor/service/agent/keys/models/usage/logs/update/code/config); TUI launcher on no-args; LaunchTarget dispatch
  • install.sh — service binary download step: downloads maschina-services-<os>-<arch>.tar.gz to ~/.local/share/maschina/bin/; graceful fallback if not in release

Fixed (2026-03-07 — CLI compilation)

  • packages/cli/src/commands/login.rs — Config struct initializers now include db_url and model_providers fields
  • packages/cli/src/config.rs — default Config in load() now includes db_url: None, model_providers: vec![]
  • packages/cli/src/commands/agent.rs — AgentRun struct now has started_at: Option<String>

Added (2026-03-07 — Session N+1)

  • Kill commands in root package.json: kill:api, kill:gateway, kill:realtime, kill:runtime, kill:daemon, kill:all
  • "type": "module" added to all TS packages missing it (ESM consistency)
  • Dotenvy env loading in services/gateway, services/realtime, services/daemon Rust services

Added (2026-03-07 — Session N: shared packages + app wiring)

  • packages/api-client — shared fetch client (api.get/post/patch/put/delete), ApiError, token helpers (localStorage); base URL from VITE_API_URL
  • packages/query — TanStack Query hooks: useAgents, useAgent, useAgentRuns, useCreateAgent, useUpdateAgent, useDeleteAgent, useRunAgent, useLogin, useRegister, useLogout, useForgotPassword, useResetPassword, useSubscription, useCredits, useCreateCheckout, useCancelSubscription, useKeys, useCreateKey, useRevokeKey, useUsageSummary, useUsageEvents, useMe, useUpdateMe, useUsers
  • packages/ui/src/index.ts — full barrel export of all 55 shadcn components
  • All 6 web apps (app, auth, admin, console, developers, web) — added @maschina/api-client and @maschina/query workspace deps
  • All 6 web apps — src/lib/api.ts replaced with re-export from @maschina/api-client (eliminates duplication)
  • All 6 web apps — __root.tsx wrapped with <TooltipProvider> from @maschina/ui
  • apps/docs — switched from Docusaurus to Mintlify; mint.json + introduction.mdx + full doc structure
  • .github/workflows/deploy.yml — fixed invalid push_tag: event (moved tags: ["v*"] under push:)
  • All 6 web apps — tsconfig.app.json + tsconfig.node.json — added "composite": true; removed deprecated "baseUrl": "."
  • packages/tsconfig/package.json — added .json-suffixed exports to fix @maschina/tsconfig/node.json resolution
  • apps/*/src/routeTree.gen.ts — stub files for TypeScript compile before first pnpm dev
  • apps/desktop/src-tauri/icons/ — full icon set: icon.icns, icon.ico, Windows Store tiles, 32/128/256px PNGs
  • apps/mobile/android/ — adaptive icons (foreground PNG + black background XML), all mipmap densities
  • apps/mobile/android/wear/ — Wear OS module: MainActivity, WearViewModel, WearApp, WatchStore, Compose UI
  • apps/mobile/ios/MaschinaWatch/ — watchOS app: WatchStore (WCSession), agents list, status view, 5 complication families

Added (2026-03-06 — Session 2)

  • packages/db/src/schema/pg/ — full PostgreSQL schema (20 modules): enums, users, auth, organizations, plans, subscriptions, api_keys, usage, credits, agents, audit, jobs, billing_events, compliance, webhooks, notifications, connectors, marketplace, misc, index
  • packages/db/src/schema/sqlite/ — full SQLite mirror schema (8 modules): users, auth, plans, subscriptions, api_keys, usage, agents, compliance, index; all pg-incompatible types replaced (sqliteTable, text enums, integer timestamps/booleans, text UUIDs, text JSON)
  • packages/db/src/client.ts — dual-dialect client: auto-detects file: prefix → SQLite (WAL + FK + busy_timeout), else → PostgreSQL (ssl for neon.tech, connection pooling)
  • packages/db/drizzle.config.ts — dialect-aware: points to pg/ or sqlite/ schema dir, separate migrations/pg and migrations/sqlite outputs
  • packages/db/src/index.ts — exports pg schema as canonical types; named re-exports for pgSchema and sqliteSchema
  • packages/db/src/rls/policies.sql — full Row Level Security: 3 DB roles (maschina_app, maschina_readonly, maschina_migrate), RLS enabled on 24 tables, user isolation policies via app.current_user_id session variable, BYPASSRLS for migrate role
  • packages/db/src/seed/plans.ts — idempotent plan seed using onConflictDoUpdate on tier; imports from @maschina/plans (canonical definitions)
  • packages/db/src/seed/index.ts — seed runner
  • packages/auth/src/types.ts — UserRole, PlanTier, AuthContext, JwtPayload, TokenPair types
  • packages/auth/src/jwt.ts — HS256 access tokens (15min), refresh tokens (30d) with separate secret, createTokenPair, generateSecureToken, hashToken
  • packages/auth/src/password.ts — argon2id (64MiB/3iter/4para), validatePasswordStrength, needsRehash
  • packages/auth/src/api-key.ts — msk_live_/msk_test_ prefix format, SHA-256 hash, timingSafeEqual comparison
  • packages/auth/src/session.ts — createSession, rotateSession (delete-on-use prevents reuse), revokeSession, revokeAllSessions, pruneExpiredSessions
  • packages/auth/src/validate.ts — resolveAuth: auto-detects API key vs JWT from Bearer header
  • packages/auth/src/rbac.ts — ROLE_HIERARCHY, PLAN_HIERARCHY, planFeatures, requireRole, requirePlan helpers
  • packages/auth/src/verification.ts — email verification (24h), password reset (1h) single-use tokens
  • packages/auth tests — jwt.test.ts, password.test.ts, api-key.test.ts, rbac.test.ts
  • packages/plans/src/types.ts — PlanTier, PlanLimits, PlanFeatures, PlanConfig, QuotaKey, QuotaStatus types
  • packages/plans/src/definitions.ts — canonical plan configs (Free/Operator/Pro/Enterprise) with token budgets; PLANS record, getPlan, isValidTier
  • packages/plans/src/gates.ts — hasFeature, isAtLeastTier, nextTier, can.* helpers, getUpgradeHints
  • packages/plans/src/quotas.ts — getQuotaStatus, withinQuota, canConsumeQuota, getAllQuotas, QUOTA_LABELS, formatLimit
  • packages/plans/src/training.ts — GDPR/CCPA training consent config: TRAINING_POLICY_VERSION, consent text, jurisdiction list, per-tier default consent, TrainingDataConfig, PII_CATEGORIES, data retention policy
  • packages/validation/ — new package: Zod schemas + sanitization + output projection
    • src/sanitize.ts — sanitizeText, sanitizeSlug, sanitizeFilename, sanitizeUrl, enforceLength; server-side only, no DOM
    • src/project.ts — projectUser, projectApiKey, projectSession, projectAgent, projectSubscription; explicit field allowlists, sensitive fields never exposed
    • src/parse.ts — parseBody, assertValid, ValidationError; safe Zod wrapper with FieldError formatting
    • src/schemas/auth.ts — RegisterSchema, LoginSchema, ChangePasswordSchema, ResetPasswordSchema, OAuthCallbackSchema
    • src/schemas/user.ts — UpdateProfileSchema, CreateApiKeySchema, UpdateTrainingConsentSchema, RequestDataExportSchema, DeleteAccountSchema
    • src/schemas/agent.ts — CreateAgentSchema, RunAgentSchema, CreateConnectorSchema, CreateWebhookSchema
  • CLAUDE.md — updated with strict session protocol: mandatory CHANGELOG + decisions.md updates during session
  • .claude/settings.json — added PostToolUse hooks that write to .claude/change-audit.log on every Write/Edit

Added (2026-03-06 — Session 2, continued)

  • services/api/ — Hono HTTP server (Node.js)
    • src/env.ts — Zod env validation, fails fast on startup with clear error
    • src/context.ts — typed Hono context: RequestUser (id/email/role/tier/sessionId/apiKeyId), Variables
    • src/app.ts — Hono app factory with global middleware + all route mounts (exported for tests)
    • src/index.ts — Node.js server entry, graceful SIGTERM/SIGINT shutdown, closes Redis cleanly
    • src/middleware/auth.ts — requireAuth (JWT + API key), optionalAuth, requireRole, requireFeature
    • src/middleware/quota.ts — requireQuota(type), trackApiCall (applied to all authed routes)
    • src/middleware/ratelimit.ts — Redis sliding window: authRateLimit (10/min), apiRateLimit (300/min), strictLimit (5/5min)
    • src/middleware/error.ts — global error handler: ValidationError→400, QuotaExceededError→429, unknown→500
    • src/middleware/cors.ts — configurable CORS, strict in production
    • src/routes/health.ts — GET /health, GET /ready (DB + Redis liveness)
    • src/routes/auth.ts — register, login, refresh, logout, verify-email, forgot-password, reset-password
    • src/routes/users.ts — /users/me profile, sessions list/revoke, training consent, GDPR export, account deletion
    • src/routes/agents.ts — agents CRUD + POST /agents/:id/run (quota enforced, dispatches to daemon)
    • src/routes/keys.ts — API key CRUD (full key shown once on creation, never again)
    • src/routes/usage.ts — GET /usage (current period quotas), GET /usage/history (paginated events)
    • src/routes/billing.ts — subscription info, checkout, plan change, cancel, portal, balance, top-up, history
    • src/routes/webhooks.ts — POST /webhooks/stripe (raw body, signature verified before processing)
    • src/jobs/reconcile.ts — nightly Redis→PostgreSQL usage reconciliation, standalone runnable
  • packages/billing/src/pricing.ts — per-action pricing rates in cents ($2.00/1M tokens, $0.01/agent run, $0.10/GB/month), calculateCost(), TOPUP_OPTIONS ($5/$10/$20/$50/$100), MIN_TOPUP_CENTS ($5 minimum matching Anthropic)
  • 7 plan tiers finalized: Access / Mach-1 (M1) / Mach-5 (M5) / Mach-10 (M10) / Teams / Enterprise / Internal
    • Access: $0 — free, local Ollama only, onboarding tier
    • M1: $20/mo or $204/yr ($17/mo) — entry paid, Maschina model unlocked
    • M5: $60/mo or $600/yr ($50/mo) — main individual tier
    • M10: $100/mo or $995/yr (~$83/mo) — power user, compliance + priority support
    • Teams: $20/seat/mo or $204/seat/yr ($17/seat/mo)
    • Enterprise: custom pricing, contact sales
    • Internal: $0, unlimited, Asher + team only, never shown publicly, no billing/quota checks
  • Prepaid balance model: $5 minimum top-up, depletes by cent, rolls over indefinitely, manual refund button (no auto-refund)
  • packages/billing/ — full Stripe billing integration
    • src/types.ts — BillingInterval, SubscriptionStatus, CreditPackage, CREDIT_PACKAGES (1M/5M/20M token packs), CheckoutResult, PortalResult
    • src/client.ts — Stripe singleton (telemetry disabled)
    • src/customers.ts — getOrCreateStripeCustomer (lazy creation, stored in subscriptions table), updateStripeCustomer
    • src/subscriptions.ts — createSubscriptionCheckout (Stripe Checkout hosted page), changeSubscriptionTier (immediate proration), cancelSubscription (at period end only), createPortalSession (Stripe Customer Portal for self-serve)
    • src/credits.ts — createCreditCheckout, getCreditBalance, addCredits (ledger + balance in DB transaction), consumeCredits
    • src/webhooks.ts — constructWebhookEvent (signature verification), handleWebhookEvent (idempotent via stripeEventId unique), routes: subscription.created/updated/deleted, invoice.paid/payment_failed, checkout.session.completed
  • packages/cache/ — Redis abstraction over ioredis: client singleton with reconnect backoff, typed ops (get/set/del/incr/decr/getJson/setJson/pipeline/multi/publish), TTL helpers (secondsUntilEndOfMonth, currentMonthKey)
  • packages/usage/ — full usage metering and quota enforcement
    • src/types.ts — UsageEventType, RecordUsageInput, QuotaCheckResult, UsageSummary, RateLimitHeaders
    • src/period.ts — UTC calendar-month period helpers (getCurrentPeriod, getPeriodForDate, secondsUntilPeriodEnd)
    • src/keys.ts — Redis key schema: quota:{userId}:{type}:{YYYY-MM}, session/plan cache keys
    • src/quota.ts — checkQuota (Redis → PostgreSQL cold-start fallback), incrementQuota (atomic INCRBY + expire), getUsageSummary, buildRateLimitHeaders (X-RateLimit-* headers)
    • src/record.ts — recordUsage (Redis incr + PostgreSQL fire-and-forget), recordModelInference (input+output tokens from model response), recordAgentExecution, recordApiCall
    • src/storage.ts — getStorageUsageGb (PostgreSQL snapshot, not Redis), checkStorageQuota
    • src/reconcile.ts — reconcileUserUsage, reconcileAllUsers (nightly Redis→PostgreSQL checkpoint sync)
    • src/middleware.ts — QuotaExceededError (429), enforceQuota, enforceAndRecordApiCall

Changed (2026-03-06 — Session 2, continued)

  • packages/db/src/schema/pg/usage.ts — added inputTokens/outputTokens columns to usage_events; changed rollup index to uniqueIndex (required for ON CONFLICT DO UPDATE upserts); updated comment to reflect Redis-first architecture
  • packages/db/src/schema/sqlite/usage.ts — added inputTokens/outputTokens columns to match pg schema

Fixed (2026-03-06 — Session 2)

  • packages/auth/src/session.ts — pruneExpiredSessions: changed gt(new Date(), sessions.expiresAt) → lt(sessions.expiresAt, new Date()) (correct Drizzle column-first argument order)
  • packages/plans/src/index.ts — corrected .ts extension to .js in ESM import

Changed (2026-03-06 — Session 2)

  • Plan monthlyModelInferences (call count) → monthlyModelTokens (token budget): Free=0, Operator=500k, Pro=5M, Enterprise=-1 (unlimited). Aligns with how all real AI platforms meter usage.

Added (2026-03-06 — Session 1)

  • Full monorepo scaffold: pnpm + Turborepo + polyglot (TypeScript, Python, Rust)
  • Rust workspace: daemon, gateway, realtime, CLI, code tool, SDK
  • Python packages: runtime (FastAPI), agents, ml (PyTorch), risk
  • TypeScript packages: types, core, db (Drizzle), auth, billing, plans, usage, keys, payments, cache, events, jobs, storage, notifications, webhooks, ratelimit, flags, errors, search, crypto, chain, compliance, reputation, marketplace, treasury, connectors, telemetry, config, content, ui, sdk/ts, sdk/python
  • TypeScript services: api, analytics, email
  • Rust services: daemon, gateway, realtime
  • Python services: worker (Celery)
  • Solana Anchor programs: treasury, marketplace
  • Docker Compose: PostgreSQL 17 + Redis 7
  • Biome for TypeScript/JSON linting and formatting
  • Vitest for TypeScript testing, pytest for Python, cargo test for Rust
  • GitHub Actions CI: TypeScript + Rust + Python jobs
  • Local CI via git hooks (pnpm hooks:install)
  • Comprehensive root package.json with 100+ pnpm commands
  • Drizzle ORM with dual-dialect support (PostgreSQL + SQLite)
  • Billing model scaffold: plans (Free/Operator/Pro/Enterprise), metered usage, API keys
  • CLAUDE.md for persistent AI session context (gitignored)
  • .claude/ folder: context.md, decisions.md, session.md for full audit trail
  • .env.example with all environment variables documented
  • All TS package stubs: tsconfig.json, src/index.ts, vitest.config.ts (34 packages)
  • All Python package stubs: src/maschina_*/__init__.py
  • All Rust stubs: src/main.rs (daemon, gateway, realtime, cli, code), src/lib.rs (sdk/rust)
  • Rust SDK: MaschinaError, Agent, AgentType, AgentStatus types
  • CLI: maschina init, agent, keys command structure (clap)
  • Root: vitest.workspace.ts, rust-toolchain.toml, .python-version, pyproject.toml
  • packages/model: GLM-4 fine-tuning scaffold (train, eval, infer, data, config)
  • packages/db schemas: users, plans, subscriptions, api_keys, usage_events, usage_rollups, credit_transactions, credit_balances, agents, agent_runs
  • packages/db client: dual-dialect PostgreSQL/SQLite with Drizzle