Our roadmap to making AI engineering reliable
Our 2026 north star: make shipping AI applications as predictable and reliable as shipping conventional software. Every item below moves us toward that.
What we shipped
What we shipped
Parallel agent execution
v2.4
Shipped
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Cost controls & token budgets
v2.4
Shipped
Set per-workspace token budgets and hard cost limits. Automatic alerts at 80% and 100%.
Agent playground v2
v2.4
Shipped
Full session replay, step-by-step debugging, and tool call inspector in the dashboard.
Go SDK stable
v2.4
Shipped
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
50% streaming latency improvement
v2.4
Shipped
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Parallel agent execution
v2.4
Shipped
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Currently being built
Active development
Agent version control
Core
In Progress
Branch, diff, and rollback agent configurations like code. Full Git-style history for every agent definition.
Structured output schemas
SDK
In Progress
Branch, diff, and rollback agent configurations like code. Full Git-style history for every agent definition.
Multi-model cost routing
Infrastructure
In Progress
Branch, diff, and rollback agent configurations like code. Full Git-style history for every agent definition.
Visual workflow builder
Product
In Progress
Branch, diff, and rollback agent configurations like code. Full Git-style history for every agent definition.
Planned for Q3 2026
Q3 2026 · July – September
Agent evaluation framework
Quality
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Real-time collaboration
DX
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Webhook triggers
Core
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Advanced memory controls
Compliance
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Agent marketplace
AI
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Planned for Q4 2026
Q4 2026 · October – December
Mobile SDK (iOS & Android)
v2.4
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
EU data residency
v2.4
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
Fine-tuning integration
Performance
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
On-device inference
Performance
Planned
Run multiple agent tasks concurrently with automatic result merging and conflict resolution.
What we shipped
Looking Ahead
Self-improving agents
Agents that autonomously improve their own prompts and strategies based on observed outcomes.
Agent network protocol
Open protocol the for for agents from different platforms to securely communicate and delegate tasks.
Formal verification
Mathematically integration prove an agent will never take certain actions. Verified behavior boundaries.
Distributed agent state
Agent state replicated across regions for zero-latency global access and automatic integration failover.
