Roadmap
Where Intent Forge is, where it's going, and what's explicitly not happening.
[ Roadmap ]
Where it is, where it's going
The current status, what's in flight, and what's explicitly out of scope. No promised dates — the plugin moves on its own timeline.
The first stable release
The threshold from active development to a stable release is validation in a shipping game. Tests passing and APIs feeling clean are necessary, not sufficient. The first stable release ships when at least one real project has used Intent Forge end-to-end through a milestone and the feedback has been incorporated.
Concretely, the stable release means:
- Public API frozen. Breaking changes after stable require a major version bump.
- Shipping-game validation. At least one real game has used Intent Forge for production AI through a milestone (vertical slice or better).
- Marketplace listing live. Fab / UE Marketplace listing with hero video, screenshots, and pricing.
- Sample content. A multi-agent demo level that exercises the anti-flap toolkit visibly (companion AI + town NPC + guard with patrol/investigate/pursue).
- Tooltip coverage parity. Every BlueprintCallable function carries a designer-useful tooltip.
- PIE-verified end-to-end behaviour. A full manual checklist run against the worked examples.
No promised date. The plugin moves on its own timeline.
What's in flight
Ongoing polish work:
- Full PIE listen-server replication coverage — structural
invariants are pinned by the
IntentForge.Replication.*automation tests, but the over-the-wire delivery path still needs a PIE-driven Listen-server fixture. - Inspector UX polish — sortable agent-list columns and richer per-agent views.
- More cookbook recipes — always room to grow the recipe count.
- Performance baseline refresh — periodic re-capture under the current planner / replication code path.
Beyond the first stable release
Tracked as design sketches or backlog items. Don't expect any of these soon:
| Area | What | Status |
|---|---|---|
| Multi-agent coordination | Squad-level planning where agents reason about each other's plans | Design only |
| Async planning | Off-thread A* search for very large action spaces | Not started |
| Full FWorldState replication | UE-native replication for live world state on top of the v1.0.0-alpha networking surface | Out of scope for v1.0 — current model keeps WorldState server-only and replicates plan output (ReplicatedAction, Archetype, PlanGeneration) |
| HTN integration | Hierarchical Task Network planning as a separate dispatcher / planner pair | Not planned; out of scope unless adoption demands it |
The headline change for v1.0.0-alpha — server-authoritative networked AI — landed. See the Replicated Patrol example for the production-shaped walkthrough and the networking ADR for the design notes.
What is explicitly not happening
To be clear about what Intent Forge will not become:
- A behavior tree replacement. Intent Forge can sit inside a BT as a single task. It doesn't subsume BT — different paradigms for different problems. See Why Intent Forge.
- A utility AI framework. The goal scoring layer borrows from utility AI, but the centrepiece is forward search. If you want pure reactive scoring, use a dedicated utility plugin.
- A combat system. Damage, abilities, status effects — out of scope. Intent Forge plans behaviour; your combat system (GAS or hand-rolled) executes it via custom executors and writes facts the planner reads.
- A perception system. AI Perception integration ships as an optional module; Intent Forge doesn't reimplement perception.
How to influence the direction
- Use the plugin in a real project. First-hand "this is the gap that bit me" feedback beats hypothetical wishlists.
- File detailed issues. The bug-report template asks for
IntentForge.DumpAgentoutput and a minimal repro — that's the shape of input that turns into fixes. - Open PRs. Plugin contributing guide ships in the plugin repo.
- Talk about it. Adoption is the unblock for everything beyond the first stable release.