Building in the Open: Our Roadmap and What's Next

We believe in building in the open. Here’s a transparent look at where we are and where we’re headed.

Updated 2026-05-16: this post is kept living. The “Shipped” section below reflects current state; items that were “in progress” or “next” when first published have either landed or moved up the stack.

What We’ve Shipped

Simple Container — Infrastructure Platform

Intent-based infrastructure primitives replacing low-level configuration
Parent Stack architecture for managing 100+ production environments
90% reduction in config complexity

Forge — AI Workflow Engine

Workflow engine: describe a team in plain English (any language) — Forge generates a workflow spec of personas + transitions + a management agent, waits for operator approval, then runs it
Management-agent pattern: a designated persona validates every handoff with a verdict — advance, enrich (loop back), or escalate — so bad output never flows downstream
Five-role SDLC pipeline (our flagship preset): PM → Architect → Developer → QA → Reviewer, all running end-to-end on the engine. Label a GitHub issue, get a reviewed PR
Review-fix rounds: bounded reviewer→developer loops, configurable per-pipeline. The proven precursor to the generic management-agent pattern above
Character agents: named personas with archetype, 10-stat character sheet, XP, levels, accumulated memory across runs. Operators can compose teams from a roster of distinct characters
Compete mode: drop two team IDs on the same brief, Forge runs two parallel pipelines side-by-side, you pick the better output
Runtime registry + capability matching: each persona declares what it needs (docker? persistent disk? short or long-running?); the matcher picks the cheapest runtime that fits, per turn
Auto-memory extractor: turns end-of-run reflections into per-character memories that carry into the next run
forge-cli: local development with bidirectional cloud sync and offline retry queue
Multi-provider AI routing: Anthropic, OpenAI, Google, Azure, Bedrock — priority + weight routing, automatic failover, per-session pinning
Per-org cloud quotas: usage metering and enforcement live
Persistent sessions: snapshots, transcripts, and full audit trails — every persona’s handoff committed to git

What’s In Motion

Compliance posture

SOC 2 Type I prep on the global track. SLSA / NIST SSDF / OWASP framework alignment ongoing.

Workflow engine UX polish

The engine is shipped; the operator-facing authoring UX (visual spec editor, prompt-history view, per-transition tuning) is the next round of work. Today’s authoring is via the workflow API + JSON spec.

Domain-aware fine-tuning

The research pipeline that trains a domain-skilled character on a company’s existing documentation + tickets + wiki. First target: industrial automation software. Outputs both a fine-tuned LoRA adapter and a domain-aware character that drops into any workflow.

On the Horizon

Marketplace for characters and presets

Public registry where operators share, fork, and remix characters + workflow presets across organizations. The 5-role SDLC preset becomes one of many; teams can ship “blog-content-pipeline-v3” or “weekly-research-summary” the same way they share code.

Sovereign-market deployments

On-prem and sovereign-cloud variants for regulated industries. Same engine, customer’s own infrastructure.

Cross-team character learning

A senior character’s accumulated memory becomes a teaching corpus for a junior character on the same or adjacent team. Learning compounds across the roster, not just within one character.

Get Involved

We’re building in the open on GitHub . If you’re interested in multi-agent AI workflows, deterministic orchestration, or autonomous infrastructure — come join us.