Reasoning

Plans like a strategist. Grounded like a lawyer.

The Hierarchical Reasoning Module plans abstractly, executes concretely, and decides when to stop. Neurosymbolic fusion grounds every embedding in a symbolic path — hallucinations become a hardware problem you can fix.

Start free Read the docs
hrm://reasonlive
goalWhy did Q3 revenue dip?
H-module · decomposition
Decompose the revenue question
Pull revenue by channel · Q3
Diff against the Q2 baseline
4Localize the mix shift
5Form causal hypothesis
6Draft recommendation
L-module · execute
$ vector.recall(similar dips, k=8)
→ matches Q3-2024 pattern
Confidence0.94
ACT · depth 4 / 6 · halts at 4neurosymbolic blend 0.6
Capabilities

What it does.

Every capability is production-grade. No flags, no betas — these ship the day you adopt Minds.

01
H-Module · abstract planner
Decomposes the goal, orders subgoals, decides when ACT stops.
02
L-Module · concrete executor
Tool calls, KG lookups, vector recall, chain-of-thought.
03
Adaptive Computation Time
The model decides when it has thought enough.
04
Neurosymbolic fusion
Neural similarity blended with symbolic graph paths per query.
05
Microtheory scoping
Reason inside a worldview — finance, medical, legal — without crosstalk.
06
Causal discovery
BFS, Dijkstra, A* path-finding over the knowledge graph.
API

A few lines is all it takes.

The SDK reflects the conceptual model directly. No glue code, no orchestrator, no learning curve beyond the data shape.

exampletypescript
const plan = await minds.reason({
  goal: "Why did Q3 revenue dip?",
  mode: "neurosymbolic",
  blend: 0.6,            // 0=pure symbolic, 1=pure neural
  microtheory: "finance",
  maxDepth: 6,           // HRM ACT cap
});

console.log(plan.steps);       // hierarchical plan
console.log(plan.confidence);  // 0.94
console.log(plan.path);        // KG path: revenue → channel-mix → ...
Specs

What it costs, what it guarantees.

Performance
Backend
Burn (Rust ML)
Models
NeoBERT 250M + Qwen3-0.6B (~2.5 GB)
Plan latency
<420ms for depth-6 plans
Path-finding
BFS · Dijkstra · A*
Counterfactuals
first-class API
Guarantees
Durability
fsync per WAL commit
Isolation
MVCC · snapshot
Audit
BLAKE3 Merkle chain
Encryption
AES-256-GCM per-namespace
Concurrency
100K+ ops/sec
Keep exploring
Knowledge Graph
Where symbolic paths live.
Cognitive Cycle
Reasoning is one stage of the loop.
Continual Learning
Reasoning skills carry across sessions.

Build it on Minds.

Start with the SDK. Ship in an afternoon. Run anywhere — Akasha Cloud, on-prem, or air-gapped.

Start free View pricing