Skip to main content
← Back to home

Changelog

New features, improvements, and fixes across AINative Studio, ZeroDB, and the AI Kit ecosystem.

v3.8.0Agent Cloud

Investor Deal Room — Agent-Accessible SAFE Round

Agents can provision a 24-hour session, read tiered deal data via MCP tools, and accept a machine-readable NDA. IEEE-7012-2025 myterms enforced on every provisioning call.

  • POST /api/v1/public/instant-db with dealroom:true provisions a session with api_key, project_id, claim_url, and 24h expiry
  • Tiered MCP access: Tier 0 (exec summary, team, traction), Tier 1 (ask, financials, cap table, deck), Tier 2 (full P&L, OpenCap — NDA required)
  • MYTERMS_ACCEPTED agreement event emitted on every dealroom provisioning call (DEALROOM-DATA-ACCESS-v1, IEEE-7012-2025)
  • POST /api/v1/public/dealroom/nda-accept for machine-readable NDA sign-off before Tier 2 access
  • ax-dealroom-mcp@0.1.0 published to npm — 12-tool MCP server for agent due diligence
  • ax-dealroom@0.1.0 CLI published to npm — provision, init, and status commands
  • @ainative-studio/dealroom-nextjs@0.1.0 published — useDealroom() React hook with provision/read/acceptNda
  • Claim URL surfaced in provisioning response — agents must surface to principal before 24h expiry
v3.7.5New

IEEE 7012-2025 Consent Management (tinyMyTerms)

Machine-readable consent declarations covering five agreement types, discoverable by AI agents via /.well-known/myterms.json and a ConsentBanner on every page.

  • /.well-known/myterms.json with five IEEE 7012-2025 agreements: SD-BASE, AD-BASIC, COMM-OPT-IN, THIRD-PARTY-OPT-OUT, DEALROOM-DATA-ACCESS-v1
  • ConsentBanner component wired into root layout — dismisses to localStorage, surfaced on first visit
  • myterms <link> discovery tag in root <head> for AI crawler detection
  • myterms.json sections added to llms.txt and agents.txt
  • Privacy page links to machine-readable terms alongside human-readable policy
v3.7.2SDKs

SDK Docs & Repo Sync — Four Packages Updated

Fixed four documentation errors that caused integration failures, and synced all SDK repos to match the production API.

  • AINativeClient constructor now correctly accepts projectId option — docs updated to match
  • REST API row insert body key is row_data (not data) — all examples corrected
  • Removed ORM-style methods (.insert, .select, .update) from docs — REST-only API correctly documented
  • ApiResponse envelope {data, status, headers, meta} now documented with typed examples
  • zerodb-typescript-sdk, zerodb-python-sdk, ainative-sdks, and ainative-zerodb-memory-mcp repos synced to core
v3.7.1Agent Cloud

Agent Swarm WebSocket, A2UI Components & Instant-DB Claim Fix

Fixed three production bugs: Agent Swarm WebSocket path mismatch, A2UI placeholder components, and the instant-db claim endpoint returning 401.

  • Agent Swarm WebSocket now connects in production — fixed path mismatch and added Kong route; CSP updated to allow wss://api.ainative.studio
  • A2UI: 11 of 17 working components now properly documented; 6 debug placeholders clearly marked as in-development
  • Instant-DB claim endpoint fixed — users can now claim agent projects after signup without 401 errors
  • Cloud Observability router registered — all three endpoints now return 200 instead of 404
v3.7.0New

AI Bot Detection + Semantic Layout

Middleware now detects AI crawlers by user-agent and serves stripped semantic HTML for cleaner LLM content ingestion.

  • AI bot user-agent detection in middleware.ts — covers GPTBot, ClaudeBot, Googlebot, Anthropic-AI, and 12 others
  • SemanticLayout wrapper component deployed across key pages — serves structured article HTML to crawlers
  • Revalidation route extended to trigger ZeroDB sync on Strapi publish events
  • llms.txt updated with corrected endpoint paths for vectors, embeddings, and events
  • agent.json discovery file added at /.well-known/agent.json
v3.6.5Agent Cloud

Agent Swarm WebSocket + Kong Route

Resolved WebSocket connectivity issues in production — path mismatch fixed, Kong route added, CSP header updated.

  • WebSocket path corrected from /ws to /agent/ws on the Kong gateway
  • Kong config updated with new WebSocket route for Agent Swarm monitor
  • Content-Security-Policy updated to allow wss://api.ainative.studio in connect-src
  • ZERODB_ENCRYPTION_KEY set and encrypted credentials now recoverable after redeploy
v3.6.3Platform

GrowthBook Replaced with Native Feature Flags + Agent Bot Logging

Removed external GrowthBook SaaS dependency — feature flags now run on our own backend. AI bot visits now logged for analysis.

  • Feature flags migrated from external GrowthBook to AINative feature_flags backend endpoint
  • agent_crawl_logs table created — AI bot visits logged with user-agent, path, and timestamp
  • POST /api/v1/public/semantic-snapshot endpoint added for triggering semantic content snapshots
  • Event logging router registered — telemetry events no longer silently dropped
  • CLI bootstrap response schema fixed — CLI and backend response contract now match
v3.6.1ZeroDB

ZeroDB Cloud: Pagination, Filtering & Count API

Improved query API with proper offset pagination, column-level ID filtering, higher query limits, and a new count endpoint for large tables.

  • Offset pagination: the "offset" parameter now works correctly as an alias for "skip" in POST /query
  • ID filtering: queries like {"id": "uuid"} now match the document row ID directly instead of searching inside JSON data
  • Query limit raised from 1,000 to 10,000 rows per request for large-table analytics
  • New POST /count endpoint: get row counts with filters without fetching all data — ideal for pagination metadata
  • Column-level filtering for id, row_id, created_at, and updated_at fields
v3.6.0Agent Cloud

Agent SDKs, Marketplace, Plugins & Billing

Standalone agent SDKs for TypeScript and Python, agent marketplace with versioning and ratings, a full plugin/extension system, per-agent billing visibility, and cognitive memory decay.

  • @ainative/agent-sdk: standalone TypeScript SDK with 12 client modules (agents, tasks, memory, vectors, files, events, billing, registry, and more)
  • ainative-agent: Python SDK on PyPI with identical API surface, async-first with Pydantic v2 models
  • Agent marketplace: registry search, semver versioning, publish workflow, and 1–5 star ratings
  • Plugin/extension system: plugin registry, per-agent install/uninstall, and runtime PluginLoader
  • Per-agent billing visibility: usage tracking, budget limits with enforcement, cost alerts, and dashboard API
  • Memory auto-decay: exponential decay worker, LRU eviction per tier, and configurable decay rates
  • All 12 OpenClaw agents bootstrapped with health monitoring and auto-restart
  • npm and PyPI publish CI workflows for all SDK packages
v3.5.0SDKs

Conversation Threads & Framework SDK v2

Persistent conversation threads with full-text search and resume, plus agent hooks and composables for all four framework SDKs.

  • Conversation Threads API: create, list, search, fork, and archive persistent chat threads
  • Thread context in chat completions: pass thread_id to resume conversations across sessions
  • Full-text thread search via PostgreSQL tsvector on titles and message content
  • React SDK v2: useAgent, useTask, useMemory, useThread hooks
  • Next.js SDK v2: server-side createAgentServerClient and webhook handler for App Router + Pages Router
  • Svelte SDK v2: createAgent, createTask, createMemory, createThread reactive stores
  • Vue SDK v2: useAgent, useTask, useMemory, useThread composables with reactive refs
  • Real-time event persistence with 24-hour replay and auto-pruning
  • Registry-to-swarm integration: submit tasks using marketplace agents via registry_agent_id
v3.4.0Community

Citizen Space Santa Cruz & Site-Wide Navigation Refresh

Launched Citizen Space Santa Cruz — a cowork + hacker + community pop-up space on the Westside. Restructured site navigation and footer for clearer developer and community paths.

  • Citizen Space landing page: founding member NFTs, weekly hack nights, 1 Gig fiber, GPU servers, Mac Mini dev machines
  • Footer restructured into 6 columns: Products, Developers, Community Resources, Moonshot Labs, Company
  • Main navigation streamlined to Products, Pricing, and ZeroDB
  • Dynamic OG images now serve across all public pages via opengraph-image.tsx convention
  • Sitemap and llms.txt updated with Citizen Space and latest platform capabilities
v3.3.0Agent Cloud

Agent Cloud: Durable State & Isolate Execution

Agents now get per-agent persistent state backed by ZeroDB agent memory, plus a new lightweight isolate execution tier for faster cold starts.

  • Per-agent durable state: ZeroDB-backed persistent storage with auto-expiry and snapshot restore
  • Lightweight isolate execution tier with auto-routing and connection pool management
  • Execution tier auto-selection: lightweight tasks route to isolates, heavy workloads to full containers
  • Snapshot management API for manual backup and restore of agent state
v3.2.0MCP

MCP Discovery API & Credential Vault

Machine-readable MCP server discovery with semantic search, plus a new credential vault for managing third-party API keys within agent workflows.

  • MCP Discovery API: search available MCP servers by capability using semantic similarity
  • Credential vault: store and retrieve third-party API keys for use in agent workflows
  • Egress proxy with domain allowlist for controlled outbound agent network access
  • MCP pricing and cost estimator endpoints for usage-aware agent orchestration
v3.1.5SEO

Dynamic OG Images for All 42 Public Pages

Every public-facing page now generates branded social preview cards dynamically using Next.js opengraph-image.tsx — no more missing or generic previews on LinkedIn, Twitter, and Slack.

  • Dynamic opengraph-image.tsx deployed to all 42 pages that previously used a static fallback
  • Blog post OG images automatically pull title, description, and category from Strapi CMS
  • Event pages generate event-specific branded cards with date and location
  • OG image endpoint returns optimized PNGs under 200KB for fast social preview loading
v3.1.0Agent Cloud

Agent Trust Scoring & x402 Payment Protocol

New trust and reputation layer for the agent cloud, with human-in-the-loop approval gates and stablecoin payment support for autonomous agent transactions.

  • Agent trust scoring: reputation-based access control for MCP tool invocations
  • Human-in-the-loop (HITL) approval gates for high-risk or high-cost agent actions
  • x402 protocol integration for stablecoin-based agent payments
  • Delegation chain middleware with cascading token revocation for multi-agent hierarchies
  • Content provider revenue sharing with per-provider analytics dashboard
v3.0.5New

New AI Models: MiniMax M2.7 & NousCoder 14B

Two new models available through the chat completions API — MiniMax M2.7 via NVIDIA NIM and NousCoder 14B for code generation workloads.

  • MiniMax M2.7: large-context model available via NVIDIA NIM inference provider
  • NousCoder 14B: code-specialized model for generation, completion, and refactoring tasks
  • Both models accessible through the standard /v1/chat/completions endpoint
  • Improved error handling: upstream provider errors now surface clearly instead of empty responses
v3.0.2ZeroDB

ZeroDB Query Improvements & Vector DB Landing Page

Fixed query filter handling in ZeroDB and launched a new landing page for the vector database use case.

  • Fixed ZeroDB query filters that were being silently ignored in certain edge cases
  • New /vector-database-for-ai-agents landing page with use-case-driven content
  • Sandbox code execution: Python, JavaScript, and TypeScript now run in isolated containers
  • Sandbox billing: per-execution billing with plan-based limits for Free, Pro, and Enterprise tiers
  • Node.js 20 runtime added to sandbox for server-side JS/TS execution
v3.0.0Platform

Platform v3.0: Agent Onboarding & AX Optimizations

Major platform update with improved agent onboarding, expanded discovery files, and comprehensive dashboard improvements.

  • New /getting-started/agents quickstart: 60-second MCP setup with runnable code examples
  • sdks.txt discovery file: machine-readable SDK directory for AI crawlers
  • PWA icons and manifest improvements for installable web app experience
  • Dashboard data pipeline: Sessions, Storage, AI Usage, Earnings, and MCP Hosting show live data
  • Conversion tracking wired across signup, events, and campaign pages
v2.8.5Knowledge Graph

Context Graph: GraphRAG & Knowledge Graph for ZeroDB

New knowledge graph capabilities for ZeroDB with ontology-aware entity resolution, edge versioning, and graph templates for common use cases.

  • Context Graph API: create entities, edges, and traverse relationship chains
  • GraphRAG: hybrid vector + graph search for more accurate retrieval
  • Knowledge Graph visualization in the ZeroDB Console
  • Graph templates for common patterns: user profiles, project dependencies, agent memory
  • Ontology support with edge versioning for structured knowledge representation
v2.8.2ZeroMemory

ZeroMemory: 100% Recall@1 on LongMemEval

ZeroMemory achieved 100% Recall@1 on the LongMemEval benchmark with integrated vector similarity search and improved memory scoring.

  • Vector similarity search integrated into ZeroMemory remember and recall flows
  • LongMemEval benchmark harness: 94% QA accuracy and 100% Recall@1 (oracle condition)
  • Improved memory scoring accuracy for temporal decay and recency weighting
  • Recall results now include full metadata for richer agent context
v2.8.0Enterprise

Enterprise Security Suite & Compliance Dashboard

Enterprise-grade security controls with expanded audit logging, compliance reporting, and fine-grained access control.

  • Compliance dashboard for SOC 2 and GDPR audit trails
  • Role-based access control (RBAC) with fine-grained scopes for API key management
  • Organization-level SSO: SAML 2.0 and OIDC with JIT provisioning
  • Audit export API for structured log export to external SIEM systems
v2.7.5ZeroMemory

ZeroMemory v2: Cognitive Memory for AI Agents

ZeroMemory now supports semantic recall, temporal decay scoring, and graph-based memory relationships — giving agents persistent, intelligent memory across sessions.

  • Hybrid vector + keyword retrieval via /memory/v2/recall
  • Temporal decay scoring: memories deprioritize over time unless reinforced
  • Memory graph API: relate memories to entities and traverse relationship chains
  • Reflection endpoint: agents summarize and compress past memories into condensed knowledge
  • Profile endpoint for per-user and per-agent preference storage
v2.7.0New

Webhook Event Dispatcher

Platform-wide webhook delivery system for real-time event notifications on agent runs, memory updates, and billing events.

  • Webhook endpoints configurable per organization from the developer settings dashboard
  • Delivery retries with exponential backoff (up to 5 attempts)
  • HMAC-SHA256 payload signing for secure delivery verification
  • Webhook delivery logs available in the developer dashboard for the past 30 days
v2.6.5Developer Program

Echo Developer Program: Usage-Based Revenue

Developers can now monetize their apps built on AINative APIs through the Echo Developer Program.

  • Set markup (0–40%) on API usage by your end users
  • AINative takes a flat 5% platform fee on all developer earnings
  • Stripe Connect integration for weekly automated payouts (minimum $10)
  • Earnings dashboard with per-app and per-user usage breakdowns
  • React SDK hooks: useChat and useCredits for integration
  • Next.js SDK with server client and auth middleware for SSR apps
v2.6.0ZeroDB

ZeroDB File Storage (S3-Compatible API)

ZeroDB now includes an S3-compatible object storage layer for agent-accessible file management.

  • PUT, GET, DELETE file operations via REST API
  • Presigned URL generation for direct browser uploads
  • Per-user and per-organization storage quotas
  • Files indexed for semantic search alongside vector embeddings
  • MCP tool integration: agents can read and write files via the ZeroDB MCP server
v2.5.3Performance

Embedding Service: 16ms Inference

The embedding service re-architected around TEI (Text Embeddings Inference) for drastically reduced latency.

  • Average embedding inference down to 16ms from ~120ms
  • Batch embedding endpoint now supports up to 512 texts per request
  • Automatic model warm-up on cold starts to eliminate first-request latency spikes
v2.5.0Security

API Key Scoping & Expiry Controls

API keys can now be scoped to specific permissions and configured with automatic expiry dates.

  • Scoped keys: limit keys to specific endpoints (e.g., memory:read, embeddings:write)
  • Key expiry: set keys to expire after a defined number of days or on a specific date
  • Last-used tracking: see when each key was last authenticated
  • Key rotation endpoint: rotate a key without downtime by generating a successor key

Older entries are available in the developer documentation.

View documentationRead the blog