Instance API
The complete registered Akasha HTTP surface, organized by capability and tied to its source contracts.
The instance API is the data and compute interface of an individual Mind. It stores memory, searches knowledge, manages model workloads, and exposes the state of the Akasha engine. The platform API manages accounts, organizations, provisioning, and billing; those are separate contracts.
The product contract is a dedicated Firecracker VM-based Akasha instance on both Free and Pro. The API families belong to that same engine on both plans. This reference verifies daemon source contracts; it does not certify that the current provisioning implementation already satisfies the Firecracker deployment requirement.
This reference accounts for 246 HTTP operations across 231 paths in the current daemon source. Every operation links its registration and handler, documents declared request fields, records response assembly, and identifies service dependencies. Registered does not mean available on every deployment: service configuration, credentials, and implementation status still apply.
Start with the connection contract
- Use the instance base URL supplied by your deployment. The direct daemon defaults to
http://localhost:8080; that is a listener default, not the hosted production address. - Read authentication and authority. Hosted instance requests use signed namespace capabilities; some operator and integration routes use separate credentials.
- Check availability and implementation limits for the capability you need.
- Use the method and path shown in the reference. An SDK method name is not evidence that a corresponding daemon route exists.
curl "$AKASHA_INSTANCE_URL/v1/keyspaces" \
-H "x-akasha-capability: $AKASHA_CAPABILITY"The variables represent your configured instance URL and issued credential. Keep credentials in your application’s secret storage. This example does not mint a capability or change your instance.
Memory and knowledge
| Capability | Reference |
|---|---|
| Events, observations, and decisions | Episodic memory |
| Stored routines | Procedural memory |
| Documents and resource metadata | Resource memory |
| Sensitive values | Vault |
| Thoughts, sessions, and interactions | Cognitive memory |
| Hybrid retrieval, emissions, and memory status | Memory |
| Active context and attention | Working memory |
| Pattern recall | Associative memory |
| Relationships, graph queries, and history | Knowledge graph |
| Knowledge lineage | Lineage |
| Review and acceptance of incoming knowledge | Learning isolation |
| Auditable removal | Forgetting |
Computation and agents
| Capability | Reference |
|---|---|
| SQL and data sources | Analytics |
| Vector indexes | Vectors |
| Retrieval and symbolic rules | Reasoning |
| Cognitive cycles | Cognitive cycle |
| Training tasks and continual state | Continual learning |
| Spiking networks and federation | Spiking networks |
| Unified inference jobs and backends | Inference |
| Embedding, generation, and training | Machine learning |
| Model downloads and lifecycle | Model management |
| Text chunking | Text chunking |
| MCP tool calls over HTTP | MCP |
| Cambium projections and tool advice | Cambium |
Storage and operation
| Capability | Reference |
|---|---|
| Keyspaces, records, and primitive import/export | Key-value storage |
| Transaction lifecycle | Transactions |
| Dataspaces and replication links | Namespaces and replication |
| Attached storage | Storage volumes |
| Backups, compaction, and indexing | Maintenance |
| Health, metrics, and instance information | Operations |
| Operator-only service controls | Administration |
Other transports and reference material
- gRPC, Arrow Flight, and MCP transport boundaries
- Errors and request limits
- Exact data types
- Coverage and known gaps
The maintainable source artifacts live under minds-docs/reference/instance/: openapi.json, routes.json, coverage.json, handler-sources.json, and schema-sources.json. Regenerate them with python3 minds-docs/scripts/generate-instance-reference.py; use --check to detect source drift. Dynamic response objects remain unconstrained in OpenAPI when the source does not establish a fixed schema.