Roadmap
Our focus for the next several releases, building GAIA as the secure, local-first AI agent platform for AMD hardware.Vote on Features — React with a thumbs-up on the GitHub issue for each item to help us prioritize.
Timeline
Shipped
v0.17.2
April 2026
v0.16.0
March 2026
v0.15.4
February 2026
mcp_servers.json). gaia init --profile mcp for one-command setup. VLM profile and structured extraction API.
v0.15.3
February 2026
v0.15.2
January 2026
v0.15.0
January 2026
Agent base class, @tool decorator, DatabaseMixin, OutputHandler abstraction. Build any agent in ~50 lines. Medical Intake Agent as reference implementation. Linux support. Mintlify docs site.
v0.14
December 2025
v0.13
November 2025
MCPAgent mixin and MCP bridge server — any agent can expose tools via JSON-RPC, any agent can consume external MCP servers. Docker agent shipped as first MCP-native agent.
v0.12
October 2025
Agent into composable mixin architecture — RAGToolsMixin, ShellToolsMixin, FileSearchToolsMixin. This is the foundation all agents build on.
v0.9–v0.11
July – September 2025
FileIOToolsMixin. Eval framework with batch experiments and ground-truth comparison. LLMClientFactory for multi-provider support (Lemonade, Claude, OpenAI).
v0.1–v0.8
April 2024 – March 2025
LemonadeClient for NPU/iGPU inference on Ryzen AI, ChatSDK (renamed to AgentSDK in v0.20.0) with model-aware prompt formatting (10+ model families), WhisperASR + KokoroTTS for voice, NSIS installer, gaia CLI entry point.
v0.17.3 — RAG Reliability, Security & Website
Status: In progress — Due: April 17, 2026 Harden the RAG pipeline, ship security fixes, and launch the public website.| Deliverable | Description | Issue |
|---|---|---|
| RAG cache security | Replace pickle with JSON + HMAC verification | #768 |
| RAG thread safety | Fix concurrent index access | #746 |
| File-search alignment | Keep file count aligned with accessible results | #754 |
| GAIA website | Public site at amd-gaia.ai | #169 |
| PDF robustness | Handle password-protected and corrupted PDFs gracefully | #451 |
| Indexing progress | Visual feedback during document indexing | #591 |
| Installer fixes | BOM and logger permission fixes | #341 |
| Username bug | Support spaces in Windows usernames | #40 |
v0.18.x — Eval, Webapp Integration & API Polish
Status: Up next Replace the eval framework, make GAIA embeddable in web applications, and harden the API.v0.18.0 — Agent Eval Benchmark — Due: April 21
| Deliverable | Description | Issue |
|---|---|---|
| Replace eval framework | New eval runner, scorecard, scenario loading | #573 |
| Eval extensibility | Plugin API for custom scenarios, scorers, document types | #671 |
| Eval test coverage | Tests for public API surface | #673 |
| Eval documentation | Scenario authoring, custom scoring, CI/CD integration | #670 |
| Legacy eval cleanup | Remove ~9,200 lines of dead code | #672 |
| Docs identity clarification | Clarify GAIA’s identity and audience paths | #724 |
v0.18.1 — Webapp Integration Support — Due: April 25
v0.18.2 — API Readiness & UI Polish — Due: April 30
| Deliverable | Description | Issue |
|---|---|---|
| Connector Hub Phase 1 | Catalog UI + Obsidian smoke test | #736 |
| System prompt optimization | Reduce ChatAgent prompt from ~7,400 to ~4,000 tokens | #719 |
| MCP security hardening | Address MCP protocol vulnerabilities | #94 |
| Unbounded context fix | Fix query context growth in chat sessions | #453 |
| Document processing tests | Comprehensive RAG test suite | #456 |
| Multi-session concurrency | Support concurrent API clients | #654 |
| Code base indexing | Indexing engine for codebase search | #656 |
| MCP test framework | Tool calling reliability tests | #709 |
| Lemonade UI feedback | Model download, busy state, connection loss indicators | #588 |
v0.19.0 — Model Fine-Tuning & Agent Quality — Due: May 5
Status: Planned Improve agent response quality through model optimization. Ship installer polish (silent install, code signing, setup wizard).| Deliverable | Description | Issue |
|---|---|---|
| Model quality optimization | System prompt compression, sub-4B model eval, first SFT fine-tune | #616 |
| Eval-to-training pipeline | Convert agent eval results into GRPO training data | #666 |
| Unsloth integration | QLoRA + GRPO fine-tuning on AMD consumer GPUs | #667 |
| LoRA adapter library | Per-agent adapters with hot-swap via Lemonade Server | #668 |
| Agent decomposition | GaiaAgent (0.6B) orchestrates specialized sub-agents | #674 |
| Agent MCP Server | Task creation, agent assignment, inter-agent communication | #675 |
| Silent install mode | Unattended/silent installation for automated deployments | #614 |
| First-run setup wizard | Guided onboarding: system scan, model download, first chat | #597 |
| Pre-configured distribution | Bundled distribution package | #711 |
| macOS code signing | Apple Developer ID for notarization | #733 |
| Windows code signing | SignPath OSS for Windows installer signing | #732 |
v0.20.0 — Agent Memory & Bootstrap — Due: May 12
Status: Planned The agent remembers you across sessions, learns your preferences, and gets smarter over time.| Deliverable | Description | Issue |
|---|---|---|
| Rename ChatAgent to GaiaAgent | Core agent rename across codebase | #696 |
| MemoryStore + MemoryMixin | Persistent memory with long conversation handling (no compaction) | #542, #543, #686 |
| Configuration dashboard | Personality recipes, skills, MCP servers, tool management | #701, #687 |
| Observability dashboard | Audit trail, activity timeline, memory browser | #697 |
| Dynamic tool loading | Load tools based on conversation context via memory | #688 |
| Merge CodeAgent views | File diff and tree views from CodeAgent to GaiaAgent | #695 |
| Bootstrap onboarding | Conversational persona interview + system discovery | #556, #470 |
| Email + Calendar (MCP) | Gmail and Google Calendar via MCP servers | #660 |
| Daily briefs | Morning digest of email, calendar, news, tasks | #663 |
| Hybrid model routing | Route simple tasks to local, complex to cloud | #632 |
| BrowserToolsMixin | Web browsing via Playwright (fallback for email/calendar) | #458 |
v0.21.0 — Multimodal, Voice & Desktop Vision — Due: May 19
Status: Planned Voice-first interaction, vision, and desktop control powered by AMD NPU acceleration. Two target demographics: Consumers (free, buying Ryzen AI hardware) and small businesses (privacy-first operations on Strix Halo).| Deliverable | Description | Issue |
|---|---|---|
| Voice-first interaction (P0) | Voice/text parity — mic in Agent UI, streaming ASR, TTS playback | #702 |
| DesktopToolsMixin (CUA) | Screenshot, mouse, keyboard control | #460 |
| ComputerUseMixin | Workflow learning and replay | #544 |
| Multi-modal rendering | Display images, audio, rich content inline in chat | #540 |
| Lemonade v10 integration | Image, voice, and system tools via Lemonade | #539 |
| Vision SDK document pipeline | Multi-page document processing with VLM | #325 |
| Voice agent demo | Memos, transcription, voice queries (ASR + TTS) | #389 |
| Meeting notes capture | Live transcription with speaker diarization, structured minutes | #700 |
| Financial tracking agent | Budget monitoring, receipt processing | #664 |
v0.22.0 — App Consolidation: Single Agent UI — Due: May 26
Status: Planned Consolidate all standalone apps (Jira, Docker, Blender, Summarize, Code, EMR, SD) into the Agent UI as selectable agents. One app, one install, all capabilities.| Deliverable | Description | Issue |
|---|---|---|
| Consolidation epic | Migrate standalone apps into Agent UI | #759 |
| Jira app | Consolidate Jira standalone app | #760 |
| Docker app | Consolidate Docker standalone app | #761 |
| Summarize app | Consolidate Summarize app | #763 |
| Blender agent | Consolidate Blender agent | #764 |
| Code agent | Consolidate Code agent (gaia-code) | #769 |
| EMR agent | Consolidate EMR agent (gaia-emr) | #770 |
| SD agent | Consolidate SD agent (gaia sd) | #771 |
| Homebrew | Add GAIA to Homebrew for macOS | #767 |
v0.22.5 — C++ Framework: Production Readiness [OEM] — Due: May 26
Status: Planned Make the native C++17 agent framework ready for OEM integration. Cancellation, event callbacks, tool security policies, streaming, runtime configuration. Detailed plan: C++ Framework Docsv0.23.0 — Autonomous Agent Infrastructure — Due: June 2
Status: Planned GAIA becomes an always-on background agent that monitors, schedules, and acts proactively.| Deliverable | Description | Issue |
|---|---|---|
| Always-on agent (autonomy engine) | Heartbeat scheduler, event hooks, background service | #634 |
| System tray app | Process manager, status indicator, quick actions | #643 |
| Home Assistant (P0) | Voice-controlled smart home, 24/7 device management, automations | #705 |
| Unified communication hub | SMS + email + messaging in single prioritized feed in Agent UI | #703 |
| Messaging adapters | Signal (privacy-first priority), Telegram, Discord, Slack | #635, #693 |
| Messaging security | Restricted default tool set, input sanitization, rate limiting | #689, #690 |
| Encrypted credential vault | Secure storage for API keys and OAuth tokens | #698 |
| Email Triage Agent | AI-powered inbox management (Tier 1 use case) | #645 |
| Calendar Agent | AI-powered schedule management (Tier 2) | #662 |
| Personal finance agent | Bank aggregation, spending analysis — platform-enabled | #707 |
| Autonomous mode | Agent schedules follow-ups and acts proactively | #555, #550 |
v0.24.0 — Agent Hub, Service Integration & Onboarding — Due: June 9
Status: Planned The ecosystem layer: discover agents, install skills, manage models, bundle for OEMs.| Deliverable | Description | Issue |
|---|---|---|
| SKILL.md format specification | Skill definition with dashboard integration | #691 |
| Agent Manifest system | Declarative metadata for agent capabilities | #462 |
| Dynamic Agent Registry | Plugin discovery, replace hardcoded wiring | #463 |
| Skill marketplace | Security tiers, AMD Verified seeding | #647 |
| OpenClaw skill compatibility | Migration tool for OpenClaw SKILL.md files | #692 |
| Model Manager UI | Download, switch, update models from desktop app | #644 |
| OEM bundling framework | Pre-configuration for hardware SKUs | #648 |
| Cost savings telemetry | Show users what they save running locally | #649 |
| Personal CRM | AI-managed contact profiles with per-person tone matching | #704 |
| Small business features | Invoicing, customer management, compliance on Strix Halo | #708 |
| Teams messaging adapter | Microsoft Teams for enterprise users | #694 |
| Legal research agent | Document analysis with RAG (Tier 2) | #699 |
| Auto-install MCP servers | Detect tools and install MCP servers during onboarding | #474 |
v0.25.0 — OS Agents: MCP Servers [OEM] — Due: June 16
Status: Planned OS-level intelligence via native MCP servers for process, network, storage, power, security, gaming, productivity, and backup management.How We Prioritize
| How You Can Influence | What It Affects |
|---|---|
| Vote on GitHub issues | Feature priority |
| Share your use case | What we build next |
| Report bugs | What we fix first |
| Contribute PRs | Faster delivery |
Get Involved
Report Issues
Found a bug or have a feature request?
Contribute
Help build GAIA
Share Your Use Case
Tell us how you’re using GAIA
Updated: April 13, 2026