Aurekai

Aurekai is the operating fabric for intelligent work, built on the Akai native runtime with compatibility for the Bonfyre runtime lineage.

Aurekai is not a wrapper around LLMs. Aurekai is a native AI operations runtime exposed through MCP, LLM tool calling, workflow engines, agent frameworks, data systems, CI/CD, infrastructure templates, and IDEs — with real commands for media, models, memory, proof, commerce, telephony, network events, reasoning, publishing, and client delivery.

Every run emits proof, lineage, metering, cache, feature, and artifact metadata.

bun add -g @aurekai/runtime
akai doctor --deep

Quick start

akai --help
akai doctor --deep
akai install --user
akai dashboard
akai run recipe.json --sae-audit --semantic-cache

What Aurekai includes

• Runtime — execution, routing, queue management, workflow dispatch, and release gates
• Commerce — auth, gate issuance, metering, invoicing, ledger, CMS, and client outreach
• Intake — file ingest, media normalization, transcription, segmentation, frame extraction, and speech loop
• Memory — .akmodel, .aksae, .akfpqx, FPQ compression, FPQx alignment, SAE activation, vector search, KV cache
• Proof — Merkle graph, lineage, equivalence indexing, SBOM, checksums, and proof bundle assembly
• Reason — reasoning sessions, branching, diff, physics trajectories, and learning feedback
• Wire — telephony simulation, PCAP ingest, wire probing/reporting, MoQ relay, and network seal/eval
• Publish — brief generation, narration, deliverable packing, clip extraction, and distribution
• Substrate — capability registry, space management, time scheduling, compression, and hardening tests

The capability registry is the source of truth: registry/aurekai.capabilities.json

Public packages

| Surface | Package | |--------------|----------------------------------------------------------------------------------| | npm | @aurekai/runtime | | PyPI | aurekai | | JSR | @aurekai/sdk | | Homebrew | aurekai/homebrew-tap | | Open VSX | aurekai-workbench | | MCP | @aurekai/mcp | | Hugging Face | aurekai/model-memory, aurekai/sae-dictionaries, aurekai/fpqx-alignments |

See the latest GitHub Release for platform binaries, manifests, SBOM, and checksums.

Ecosystem integrations

Aurekai maintains active integration surfaces across LLM providers, agent frameworks, workflow orchestrators, data/ML tools, infrastructure platforms, IDE/dev environments, MCP, package managers, and model-memory registries.

Each integration is a host-native adapter over the Akai runtime — not a wrapper. Every integration declares which capability families it exposes, which native Akai commands it binds, which platform-native primitives it exploits, and which proof/lineage/metering artifacts it emits.

See registry/integrations.json and ECOSYSTEM_NAMES.md for the full matrix.

Deep integrations

Aurekai integrations are capability-native adapters over the Akai runtime, not thin shell wrappers.

See:

• registry/integrations.json
• registry/aurekai.capabilities.json

LLM providers

OpenAI · Anthropic · Gemini · Mistral · Groq · xAI · Perplexity · Cohere · local Llama stacks · Gateway · Evals

Agent frameworks

LangChain · LlamaIndex · Haystack · Semantic Kernel · AutoGen · CrewAI

Workflow and orchestration

Kestra · Airflow · Prefect · Dagster · Temporal · n8n · Node-RED · GitHub Actions · GitLab CI · Argo Workflows · Tekton

Data and ML

MLflow · W&B · DVC · DuckDB · SQLite · PostgreSQL

Infrastructure and developer environments

Terraform · Pulumi · Helm · Kubernetes · Nix · VS Code · Open VSX · Dev Containers · Codespaces · Gitpod

Capability architecture

Akai Native Runtime
  → Aurekai Capability Registry  (registry/aurekai.capabilities.json)
  → Generated tool/workflow/schema bindings
  → Deep integrations for workflow engines, CI/CD, infra, IDEs, data/ML, MCP, agent frameworks, LLM providers
  → Every run emits proof, lineage, metering, cache, feature, and artifact metadata

Capability families and their primary commands:

| Family | Key commands | |--------------|---------------------------------------------------------------------| | runtime | runtime.capabilities, control.route, queue.enqueue, tier.route, workflow.run | | commerce | gate.issue, meter.record, pay.invoice, ledger.export, outreach.followup | | intake | ingest.file, transcribe.audio, segment.speakers, frame_extract.video, speech_loop.transform | | memory | model.pull, fpq.compress, fpqx.align, sae.activate, vec.search, kvcache.chain | | proof | canon.hash, graph.lineage, graph.merkle, index.equivalence, proof.bundle | | reason | reason.start, reason.branch, physics.run, flow.branch, learn.feedback | | wire | tel.sim_call, wire.ingest_pcap, wire.report, moq.video_relay, net.seal | | publish | brief.generate, narrate.brief, pack.deliverable, clips.extract, distribute.bundle | | substrate | capability.registry, space.put, time.schedule, compress.family, violence.coupling_test |

Validation: node scripts/validate-capability-bindings.mjs

Formats

Aurekai-first

| Format | Purpose | |------------|--------------------------------------| | .akmodel | model weight pack | | .aksae | SAE feature dictionary | | .akfpqx | cross-model FPQx alignment manifest |

Legacy-compatible

| Format | Alias | |------------|------------| | .bfmodel | .akmodel | | .bfsae | .aksae | | .bffpqx | .akfpqx |

Manifest bridge

Every Aurekai release includes:

| Manifest | Purpose | |-------------------------|------------------------------------------| | aurekai.manifest.json | public distribution metadata | | bonfyre.manifest.json | native runtime compatibility validation |

This is not a bug. It is the bridge.

CLI

akai doctor --deep
akai dashboard
akai run <recipe> [--input FILE] [--sae-audit] [--semantic-cache]
akai install --user
akai sae:activate --dict default.aksae --residual residual.bin
akai model:inspect qwen3-8b.akmodel
akai fpqx:align-sae --from qwen3-8b.l24.aksae --to llama3-8b.l26.aksae --out qwen3-to-llama3.akfpqx
akai query:features "feature:6159 > 0.7"

Compatibility aliases remain during migration:

bonfyre       -> akai
bonfyre-hyper -> akai
bonfyre-sae   -> akai sae

Native runtime lineage

Aurekai is the public platform for intelligent-work automation, built on a native runtime lineage originally developed under the Bonfyre codename.

The native runtime is tracked separately: https://github.com/aurekai/native-runtime

See COMPATIBILITY.md and MIGRATION.md for migration mechanics, dual-manifest rules, and .bf* format support.

License

Aurekai code and tooling are licensed under Apache-2.0.

Model-memory artifacts, SAE dictionaries, FPQx alignments, and benchmark assets may carry additional upstream model or dataset terms. See each artifact manifest.