Notes, reports, datasets, observations, recipes, protocols, and archives.
Local-first semantic infrastructure
Build worlds from meaning, not tables.
Akasha is a semantic runtime for knowledge, workflow, and collaboration. Explore a live semantic knowledge space in your browser — then build your own with the same open-source foundation.
No installation to exploreNo registrationMIT licensedRuns locally
Semantic
RuntimeAtom · Relation · Set
SETSGRAPHWORKFLOWSOCIETYSENSORPRESENTATIONRuntimeAtom · Relation · Set
Experience it first
One engine. Multiple worlds.
Akasha is easier to understand by walking through it. These public sites run on the same semantic foundation and are available through guest sessions in an ordinary browser.
Live public system · Guest access
Akashic Archives
Enter through a world map, reference hall, exhibition room, field gallery, or the Semantic Space Tour.
Enter the archives →Domain presentation · FoodAkashic Kitchen
Explore ingredients and recipes through season, place, diet, technique, equipment, culture, and overlapping semantic sets.
Open the kitchen →The core idea
A concept model can also become an execution model.
The ontology is optimized into a runtime where graphs and pre-bundled sets can be traversed, intersected, projected, and reused directly.
Do the slow work once.
Explore at human speed.Loading, canonicalization, weaving, and set bundling prepare the semantic space in advance so interactive operations stay small.
Explore at human speed.Loading, canonicalization, weaving, and set bundling prepare the semantic space in advance so interactive operations stay small.
Atoms, typed relations, sets, provenance, time, scope, and domain meaning.
Canonicalization, weaving, and one-level set access are prepared automatically.
Set algebra, graph traversal, workflows, vector plugins, and controlled reasoning.
Browser, Swift, CLI, MCP, RPC, maps, galleries, and research tools.
Design principles
Keep meaning, action, and agency separate.
Humans, LLMs, scripts, sensors, and future devices can use one substrate without collapsing their responsibilities into one another.
Meaning stays addressable
A concept is not trapped inside a document, table, UI, or model response.
Context can overlap
Sets form reusable semantic spaces. Their intersections express combinations rigid inheritance handles poorly.
Knowledge can execute
The same dynamic graph can describe a recipe, workflow, research protocol, or Harmonia job.
Roles shape behaviour
Humans and LLMs join as clients with bounded responsibilities; Society supports role-based review.
Local-first by default
Core knowledge and deterministic operations can stay local. External models are optional participants.
Presentation stays thin
HTML, JavaScript, and Swift focus on interaction while semantic rules remain in Akasha.
From ordinary to ambitious
The same substrate can support very different practices.
Kitchen is a friendly entrance, but the underlying model combines strict sets, graph structure, optional vector proximity, and bounded LLM reasoning.
Food and nutrition
Season, equipment, cooking method, allergy exclusions, traditions, nutrition, and recipe workflows.
Research knowledge
Organize dormant reports and experiment records while preserving provenance and uncertainty.
Materials exploration
Intersect constraints, add vector-neighbour candidates, trace graph evidence, and review through specialist roles.
Protocols and production
A recipe, laboratory protocol, manufacturing process, and workflow can share one dependency graph.
Archives and curation
Build reading halls, map entrances, exhibitions, field galleries, and permanent concept pages.
Interfaces to the world
Map atoms to sensors, actuators, services, LLMs, or simulated agents through one semantic interface.
A practical research path
Start with the knowledge you already paid to create.
Many laboratories have valuable material sleeping in PDFs, spreadsheets, reports, and experiment notes. Akasha offers a path from scattered records to a local semantic platform.
Use LLMs for preparation, not permanent dependency.
Affordable models can extract terms, normalize units, propose atoms and relations, and flag ambiguous passages for human review.
Once approved and loaded, the result becomes reusable local structure. Deterministic set and graph operations no longer require sending the source material to an LLM for every question.
Suitability and cost depend on volume, quality, and validation needs. Akasha is infrastructure, not a substitute for domain review.
- Gather dormant material
Reports, tables, notes, protocols, and curated datasets. - Prepare with bounded LLM tasks
Extraction, normalization, provenance, and format conversion. - Validate with domain experts
Approve important concepts and preserve uncertainty. - Compile the semantic space
Load, weave, canonicalize, bundle sets, and optimize once. - Add only what is needed
Thin UI, vector plugin, Society review, workflow, or instrument integration.
Two interchangeable foundations
A stable default. A native path forward.
The SQLite and Silica builds share the same interface and upper layers. Choose the mature default, or evaluate the native engine without changing the application.
SQLite build
The conservative choice: a mature storage foundation with stronger concurrent write throughput and predictable tail latency.
- Best current choice for general production use
- Same concepts, sessions, workflows, and clients
- Fallback remains available at any time
Silica build
An Akasha-native engine that treats Atom, Relation, and Set operations as its own execution vocabulary rather than adapting them to SQL.
- Crash recovery and correctness match SQLite in current tests
- Competitive median latency for isolated operations
- Concurrent throughput and tail latency still need work
| Test | SQLite | Silica |
|---|---|---|
| Crash-stop (SIGKILL) | 0 data loss | 0 data loss |
| Single-operation throughput | 179.7 writes/s · p50 4.6 ms · p99 28 ms | 128.8 writes/s · p50 4.0 ms · p99 73 ms |
| Concurrent storm (1,600 writes) | 64 writes/s · p99 203 ms | 38 writes/s · p99 754 ms |
| Dedup · isolation · priority · guest pool | All pass | All pass |
Current result: equivalent correctness and recovery; good isolated median latency; weaker concurrent throughput and tail latency in Silica.
Choose your entrance
Explore first. Install when you are ready.
The browser is the lowest-friction introduction. Developers can then download a self-expanding seed and run the same architecture locally.
Explore the live system →Guest session · no installation · no registrationRead the Quick Start →Install, start, and take the first semantic-space tourOpen the Cookbook →Concept models, maps, workflows, Society rooms, and presentationsBrowse the source →MIT licensed · Python · SQLite and Silica editions
shell · example
# 1. Give Akasha an empty home $ mkdir akasha && cd akasha # 2. Download a release seed, then run it once $ python akasha_seeds_latest.py # 3. Start the expanded system directly next time $ python akasha.py # Use the current filename from Releases / Quick Start.
Do not begin with the database.
Begin with the world you want to describe.
Walk through the public archives, visit the kitchen, inspect the source, and decide which entrance fits your own knowledge.