Release notes
Changelog
Every update, improvement, and fix — documented.
v0.5.0
ReleaseThe agent pipeline is now configurable end-to-end, the desktop war room is faster and more transparent, and the admin backoffice gains live model testing plus a third sign-in method.
- Configurable agent pipeline — administrators can adjust individual stages' settings without redeployment; changes apply on the next run.
- Stage 01 clarification loop — when a request is ambiguous the council asks back instead of guessing, iterating with follow-up questions until the request is genuinely clear.
- Live model connection test — every managed AI model now has a one-click test button that fires a real round-trip and reports latency.
- Public usernames — pick a handle when you sign up; the desktop app shows it everywhere your identity surfaces.
- Diagnostics panel in the desktop app — a dedicated bottom-area view collects every workflow event, error, and request with copy-to-clipboard for fast support handover.
- Faster first response — the desktop app pre-warms the council before you press Enter, so the run starts almost instantly instead of pausing for setup.
- Long agent outputs are no longer truncated — every reply expands inline with a Show more / Show less control.
- Stage cards collapse and re-open at any time, including while the stage is still active.
- Long-running runs no longer drop on idle — the live stream keeps itself alive through corporate proxies and edge networks.
- Admin Providers and Models are now cleanly separated — connections in one place, model picks and tests in another.
- Rebranded the council picker to a simple "Models" label for clarity, with hover-prefetch so the popup opens instantly.
- Stricter token-version handling on plan and admin changes — sessions are evicted immediately when an account boundary changes.
- Per-user usage records are now written for every run outcome, including failed and cancelled ones, so quota enforcement matches reality.
v0.4.0
ReleaseInternal reliability and email deliverability improvements.
- Expanded transactional email coverage — you now get a confirmation for every account-impacting event: password change, email change, 2FA toggle, invoice receipt, account deletion, subscription activated/cancelled, team invite, and newsletter welcome/unsubscribe.
- Reply threading — outbound replies from the team now show up in the same Gmail/Outlook conversation as your original message.
- Hardened request-size handling end-to-end to reject oversized uploads at the network edge.
- Stricter session invalidation on admin privilege changes — stale sessions are cut immediately.
v0.3.0
ReleaseThe multi-agent consensus runtime, stronger session security, and faster dashboard delivery are now in place.
- Multi-agent consensus runtime — the guarded multi-step adversarial workflow runs across the managed model catalogue
- Stronger session security for account and admin sessions
- Faster dashboard delivery through infrastructure improvements.
- AI model catalogue expanded across budget, standard, and frontier tiers (managed selection per plan).
- Database reliability improvements for high-throughput sessions.
- Backend infrastructure migration completed for faster response times.
- Encryption-at-rest for all sensitive fields with a versioned key rotation scheme.
- Modern password hashing with automatic upgrade of legacy accounts on login.
- TOTP 2FA and backup codes enabled platform-wide.
v0.2.0-alpha
AlphaMajor platform update with security hardening, 2FA, team features, and new AI models.
- Two-factor authentication with authenticator app support.
- Multi-agent consensus runtime — the guarded multi-step adversarial workflow runs across the managed model catalogue.
- Code understanding — automatic import, call-graph, and inheritance detection.
- Graph-aware retrieval over your code dependency graph.
- Dashboard deployment page now hands users back to the desktop workflow instead of exposing web-side deployment controls.
- Team management — roles (admin/member/viewer) and email invitations.
- Desktop chat history stays local on the user device instead of being stored server-side.
- Per-model credit pricing so you pay proportional to what you use.
- AI model catalogue expanded across budget, standard, and frontier tiers (managed selection per plan).
- Distributed rate limiting so quotas work consistently across multiple instances.
- Agent run token usage is now metered toward credit quotas for all run outcomes.
- Pricing display now follows the backend plan catalog instead of hardcoded launch copy.
- Annual billing removed — monthly-only pricing model.
- Public pricing page simplified — technical token limits hidden.
- Homepage metrics replaced with real feature data.
- Abuse-protection limits now enforced consistently across every entry point.
- Hardcoded development secrets removed from production code paths.
- Newsletter unsubscribe links signed with a proper secret.
- Ownership checks added to job, agent, and chat endpoints.
- Admin-only endpoints no longer accessible without authentication.
v0.1.0-alpha
AlphaInitial alpha release of SAMI — the Multi-Agent AI IDE.
- Multi-agent orchestration engine — administrators configure the number of parallel council members per run.
- Multi-agent consensus orchestration — agents debate, vote, and record the accepted solution
- Managed model catalog spanning budget, standard, and frontier model tiers.
- Hybrid context engine combining semantic and keyword retrieval for codebase awareness.
- Deep code understanding — import resolution, call-graph detection, and inheritance analysis across Python, TypeScript, Rust, and JavaScript
- War Room UI — watch agents discuss and debate in real-time
- Full-featured code editor with IntelliSense and syntax help.
- Integrated terminal embedded in the IDE.
- Live Diff View for streaming code changes
- SSE and WebSocket streaming for real-time chat
- LLM Router with automatic fallback and provider abstraction
- Modular, highly concurrent backend designed for horizontal scale
- Database layer improvements for high-throughput concurrency.
- Subscription and credit management via a managed billing integration
- Rate limiting and session management
- Context retrieval understands code symbols — functions, classes, and imports are indexed for precise retrieval.
- Improved relevance ranking for codebase context retrieval.
- Encrypted desktop credential storage
- Managed model architecture for enterprise data sovereignty
