Capability catalogue · 100 items, audited
The AI data plane,
item by item.
Every capability carries the status a dated audit gave it, and links to that audit. Statuses come from a dated capability audit that names the source file, API surface and test result behind each one. The audit changed three of its own entries on review — two down, one up — and those are published as found. It is a single-operator internal audit, not third-party certification. Six entries were re-checked after the audit against engine edition 5f8ea3af and now carry a higher status than the audit gave them: retrieval policy reranking, managed embedding jobs, approval gates, data-residency policy, the local evaluation harness and labelled relevance datasets. Their evidence rows below still show what the audit found on its date.
- 54available
- 12beta
- 15roadmap
- 19research
Browse the surface area
Find the capability you need
Showing all 100 audited capabilities
BM25 full-text retrievalRetrievalavailableShow details for BM25 full-text retrieval
Keyword retrieval is a native query shape over entity attributes.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Cosine vector retrievalRetrievalavailableShow details for Cosine vector retrieval
Search embedding-bearing entities by semantic similarity without a separate vector service.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Hybrid product retrievalRetrievalavailableShow details for Hybrid product retrieval
Fuse keyword candidates, facets, and optional cosine reranking in one product-search call.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Kind-scoped searchRetrievalavailableShow details for Kind-scoped search
Constrain retrieval to a record kind before ranking to keep application domains separate.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Unicode-aware tokenisationRetrievalavailableShow details for Unicode-aware tokenisation
Normalise case and accents as text enters the native search index.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Incremental retrieval indexRetrievalbetaShow details for Incremental retrieval index
Use a persisted, incrementally maintained retrieval path for growing collections.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Predicate-aware ANNRetrievalbetaShow details for Predicate-aware ANN
Plan vector retrieval that understands structured filters before candidate selection.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Retrieval policy rerankingRetrievalavailableShow details for Retrieval policy reranking
Apply explicit ranking policies across lexical, semantic, and business signals. A policy reranks candidates your application supplies in a second request; the search routes do not call it.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Multimodal retrievalRetrievalresearchShow details for Multimodal retrieval
Retrieve over image, audio, and text representations under one query contract.
Exploratory work; not scheduled as a feature.Evidence for this status →Learned hybrid fusionRetrievalresearchShow details for Learned hybrid fusion
Learn how to combine retrieval signals from labelled relevance data.
Exploratory work; not scheduled as a feature.Evidence for this status →Provider-agnostic completionsModels & embeddingsavailableShow details for Provider-agnostic completions
Use one completion response shape across supported LLM providers.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Anthropic providerModels & embeddingsavailableShow details for Anthropic provider
Call the Anthropic Messages API through the pluggable provider layer.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →OpenAI-compatible providerModels & embeddingsavailableShow details for OpenAI-compatible provider
Use endpoints that implement the OpenAI chat-completions surface.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Local Ollama providerModels & embeddingsavailableShow details for Local Ollama provider
Run local models through an Ollama-compatible provider without a cloud key.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Provider fallback chainsModels & embeddingsavailableShow details for Provider fallback chains
Try configured model providers in a deliberate fallback order.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Embedding dimension contractsModels & embeddingsavailableShow details for Embedding dimension contracts
Keep embedding dimensions explicit at write and query boundaries.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Model routing policyModels & embeddingsroadmapShow details for Model routing policy
Route workloads to models by latency, privacy, and task shape.
Planned direction, not a current product promise.Evidence for this status →Managed embedding jobsModels & embeddingsbetaShow details for Managed embedding jobs
Generate and refresh embeddings as a first-party managed workload. A run is synchronous and uses an Ollama-compatible embedding endpoint read from the environment.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Model cost observabilityModels & embeddingsroadmapShow details for Model cost observability
Attribute tokens, latency, and spend by model and application action.
Planned direction, not a current product promise.Evidence for this status →Fine-tuning registryModels & embeddingsresearchShow details for Fine-tuning registry
Track deployable fine-tunes beside the data and evaluation records that justify them.
Exploratory work; not scheduled as a feature.Evidence for this status →Structured tool definitionsAgents & toolsavailableShow details for Structured tool definitions
Define tool inputs with JSON Schema for provider-independent calling.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Tool-call round tripsAgents & toolsavailableShow details for Tool-call round trips
Carry tool calls and results through a common message model.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Memory collectionsAgents & toolsavailableShow details for Memory collections
Store agent state in queryable collections that share the data plane.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Scoped operational toolsAgents & toolsavailableShow details for Scoped operational tools
Design operations around bounded tools rather than unrestricted shell access.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Agent audit trailsAgents & toolsavailableShow details for Agent audit trails
Keep an auditable record of agent operations and their effects.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Approval gatesAgents & toolsbetaShow details for Approval gates
Require explicit review before destructive or expensive agent actions. Held actions wait for a decision and send no outbound notification, so an operator watches the approvals list.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Durable agent runsAgents & toolsroadmapShow details for Durable agent runs
Persist resumable, long-running agent work with checkpoints.
Planned direction, not a current product promise.Evidence for this status →Agent regression suitesAgents & toolsroadmapShow details for Agent regression suites
Run task suites against agent versions before promotion.
Planned direction, not a current product promise.Evidence for this status →Multi-agent role orchestrationAgents & toolsresearchShow details for Multi-agent role orchestration
Coordinate specialist agents with clear privileges and handoffs.
Exploratory work; not scheduled as a feature.Evidence for this status →Self-improving tool selectionAgents & toolsresearchShow details for Self-improving tool selection
Learn safe tool-selection strategies from reviewed traces.
Exploratory work; not scheduled as a feature.Evidence for this status →Document entitiesData foundationavailableShow details for Document entities
Store flexible JSON entities as first-class durable records.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Key-value accessData foundationavailableShow details for Key-value access
Use direct key reads for exact context and application state.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Metadata filtersData foundationavailableShow details for Metadata filters
Filter retrieval against structured application attributes.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Files alongside recordsData foundationavailableShow details for Files alongside records
Keep file metadata and permissions aligned with the application data model.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Events in the data planeData foundationavailableShow details for Events in the data plane
Build event context from the same mutation stream as application records.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Typed schema viewsData foundationbetaShow details for Typed schema views
Expose strongly typed application views over flexible documents.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Knowledge ingestion pipelinesData foundationbetaShow details for Knowledge ingestion pipelines
Turn source files and connectors into governed retrieval records.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Document lineageData foundationroadmapShow details for Document lineage
Track source, transform, and chunk lineage for every retrieval record.
Planned direction, not a current product promise.Evidence for this status →Semantic deduplicationData foundationresearchShow details for Semantic deduplication
Detect near-duplicate knowledge before it enters a corpus.
Exploratory work; not scheduled as a feature.Evidence for this status →Temporal knowledge queriesData foundationresearchShow details for Temporal knowledge queries
Ask which facts were valid at a particular time in a changing corpus.
Exploratory work; not scheduled as a feature.Evidence for this status →Auth and API keysGovernanceavailableShow details for Auth and API keys
Use application authentication and API-key controls in the same system.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Roles and scopesGovernanceavailableShow details for Roles and scopes
Model access to data and operations through roles, keys, and scopes.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Audit eventsGovernanceavailableShow details for Audit events
Treat audit events as part of the application and operations model.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Tenant boundariesGovernanceavailableShow details for Tenant boundaries
Scope records and retrieval to the owning project or tenant.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →File permissionsGovernanceavailableShow details for File permissions
Align file access metadata with application permissions.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Retrieval policiesGovernancebetaShow details for Retrieval policies
Define which collections and fields an agent may retrieve from.
Implemented but incomplete — the audit states what is missing.Evidence for this status →PII redaction flowsGovernanceroadmapShow details for PII redaction flows
Apply reviewable redaction before model calls or exports.
Planned direction, not a current product promise.Evidence for this status →Data-residency policyGovernancebetaShow details for Data-residency policy
Choose deployment and storage boundaries that meet regional needs. A decision and drift API is reachable; it answers placement questions and does not intercept the engine.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Policy simulationGovernanceresearchShow details for Policy simulation
Preview whether an access policy changes what an agent can retrieve.
Exploratory work; not scheduled as a feature.Evidence for this status →Provenance confidence scoringGovernanceresearchShow details for Provenance confidence scoring
Score confidence from source authority, freshness, and transformation history.
Exploratory work; not scheduled as a feature.Evidence for this status →Change feedsRealtime contextavailableShow details for Change feeds
Derive application updates from the same mutation stream as records.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →SSE subscriptionsRealtime contextavailableShow details for SSE subscriptions
Subscribe to realtime updates over server-sent events.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Realtime eventsRealtime contextavailableShow details for Realtime events
Keep agents and interfaces current with data-plane events.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Durable conversation contextRealtime contextavailableShow details for Durable conversation context
Persist context as application data instead of an isolated session cache.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Local-first contextRealtime contextavailableShow details for Local-first context
Run context storage close to the application in a portable deployment.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Streaming retrieval updatesRealtime contextbetaShow details for Streaming retrieval updates
Push new evidence into an active retrieval session as it arrives.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Live agent handoffRealtime contextroadmapShow details for Live agent handoff
Transfer a working context between a person and an agent in real time.
Planned direction, not a current product promise.Evidence for this status →Event-driven agentsRealtime contextroadmapShow details for Event-driven agents
Invoke governed agent workflows from durable data events.
Planned direction, not a current product promise.Evidence for this status →Causal context graphRealtime contextresearchShow details for Causal context graph
Model the events that caused a fact or recommendation to change.
Exploratory work; not scheduled as a feature.Evidence for this status →Collective memory controlsRealtime contextresearchShow details for Collective memory controls
Control what teams and agents may contribute to shared memory.
Exploratory work; not scheduled as a feature.Evidence for this status →Open data planePortabilityavailableShow details for Open data plane
Run the core system without adopting a single cloud provider.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Single-binary deploymentPortabilityavailableShow details for Single-binary deployment
Start with one deployable database process for the data plane.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Local and Docker workflowsPortabilityavailableShow details for Local and Docker workflows
Develop locally and package the same data plane for deployment.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Cloud runtime profilesPortabilityavailableShow details for Cloud runtime profiles
Use named deployment profiles instead of exposing provider knobs first.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Provider boundariesPortabilityavailableShow details for Provider boundaries
Keep provider-specific behaviour outside the storage kernel.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Edge-read profilePortabilitybetaShow details for Edge-read profile
Offer a tuned read-oriented deployment profile for edge-facing workloads.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Portable backup bundlesPortabilityroadmapShow details for Portable backup bundles
Move data and retrieval state between compatible environments.
Planned direction, not a current product promise.Evidence for this status →Cross-cloud migrationPortabilityroadmapShow details for Cross-cloud migration
Plan low-friction movement of an AI app between cloud providers.
Planned direction, not a current product promise.Evidence for this status →Air-gapped AI deploymentsPortabilityresearchShow details for Air-gapped AI deployments
Package models, retrieval, and operations for disconnected environments.
Exploratory work; not scheduled as a feature.Evidence for this status →Sovereign control planePortabilityresearchShow details for Sovereign control plane
Operate a fleet without exporting sensitive operational metadata.
Exploratory work; not scheduled as a feature.Evidence for this status →Health contractsOperationsavailableShow details for Health contracts
Expose health checks for a portable data-plane process.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Metrics and logsOperationsavailableShow details for Metrics and logs
Operate through application metrics, logs, and explicit lifecycle commands.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Admin operationsOperationsavailableShow details for Admin operations
Manage collections, keys, users, backups, and deploy targets from familiar operations surfaces.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Backup-aware architectureOperationsavailableShow details for Backup-aware architecture
Treat backup and recovery as first-class operating concerns.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Published benchmark evidenceOperationsavailableShow details for Published benchmark evidence
Publish measurement context instead of unsupported speed claims.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Retrieval SLO dashboardsOperationsroadmapShow details for Retrieval SLO dashboards
Track latency, recall, and freshness together for production retrieval.
Planned direction, not a current product promise.Evidence for this status →Model incident responseOperationsroadmapShow details for Model incident response
Run a bounded response playbook for provider or quality failures.
Planned direction, not a current product promise.Evidence for this status →AI capacity planningOperationsroadmapShow details for AI capacity planning
Plan storage, retrieval, and inference capacity as one service envelope.
Planned direction, not a current product promise.Evidence for this status →Autonomous remediationOperationsresearchShow details for Autonomous remediation
Let reviewed agents propose or execute bounded operational repairs.
Exploratory work; not scheduled as a feature.Evidence for this status →Counterfactual retrieval debuggingOperationsresearchShow details for Counterfactual retrieval debugging
Explain which source or ranking decision changed a model outcome.
Exploratory work; not scheduled as a feature.Evidence for this status →HTTP APIDeveloper surfaceavailableShow details for HTTP API
Integrate application data and retrieval over an explicit server API.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →TypeScript clientDeveloper surfaceavailableShow details for TypeScript client
Use a TypeScript client for application-facing database interactions.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →MCP surfaceDeveloper surfaceavailableShow details for MCP surface
Connect compatible tools through a model-context protocol surface.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →CLI lifecycleDeveloper surfaceavailableShow details for CLI lifecycle
Initialize, start, and inspect a portable local data-plane process.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Rust coreDeveloper surfaceavailableShow details for Rust core
Build on a native Rust engine with explicit systems boundaries.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →SDK retrieval helpersDeveloper surfacebetaShow details for SDK retrieval helpers
Add typed helpers for common retrieval and context patterns.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Prompt versioningDeveloper surfaceroadmapShow details for Prompt versioning
Version prompts with the application code and deployment context.
Planned direction, not a current product promise.Evidence for this status →Local evaluation harnessDeveloper surfaceavailableShow details for Local evaluation harness
Run retrieval checks before pushing an application change. The harness runs in process against the engine and reports recall, NDCG and MRR; agent suites are a separate beta command.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Semantic debuggerDeveloper surfaceresearchShow details for Semantic debugger
Inspect the context and ranking path behind an individual response.
Exploratory work; not scheduled as a feature.Evidence for this status →Intent test generationDeveloper surfaceresearchShow details for Intent test generation
Generate reviewed test queries from a product’s real information architecture.
Exploratory work; not scheduled as a feature.Evidence for this status →Measured retrieval benchmarksEvaluation & qualityavailableShow details for Measured retrieval benchmarks
Publish documented benchmark context for retrieval performance.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Retrieval benchmark workloadsEvaluation & qualityavailableShow details for Retrieval benchmark workloads
Exercise BM25 and vector paths with named retrieval workloads.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Honest capability statesEvaluation & qualityavailableShow details for Honest capability states
Separate implemented, partial, and future capability claims.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Architecture quality gatesEvaluation & qualityavailableShow details for Architecture quality gates
Use explicit proof requirements before broad production claims.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Source-linked feature claimsEvaluation & qualityavailableShow details for Source-linked feature claims
Link public available-state claims to the product evidence behind them.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Recall measurementEvaluation & qualitybetaShow details for Recall measurement
Compare approximate retrieval against an exact baseline for known datasets.
Implemented but incomplete — the audit states what is missing.Evidence for this status →Labelled relevance datasetsEvaluation & qualityavailableShow details for Labelled relevance datasets
Evaluate retrieval quality with a maintained, task-specific judgement set. Provenance is recorded, not verified, and there is no judging interface.
Implemented and exercised by tests; the audit names the source file and API surface.Evidence for this status →Prompt regression trackingEvaluation & qualityroadmapShow details for Prompt regression tracking
Track answer quality changes as prompts, models, and corpora evolve.
Planned direction, not a current product promise.Evidence for this status →Human preference evaluationEvaluation & qualityresearchShow details for Human preference evaluation
Collect structured reviewer preferences for consequential outputs.
Exploratory work; not scheduled as a feature.Evidence for this status →Continuous trust modelEvaluation & qualityresearchShow details for Continuous trust model
Combine quality, freshness, source, and policy signals into a decision to answer.
Exploratory work; not scheduled as a feature.Evidence for this status →No capabilities match those filters. Try a broader search or restore a status.
Evidence · audit dated 2026-08-12
What backs each status
For every capability, the source, API surface or test the audit found — or the reason it found nothing. It is a single-operator internal audit, not third-party certification. The same record is available as JSON.
Changed on the audit's own review
- Approval gatesbetachanged toroadmap
The audit found approval records, UI, timeout and routes, but did not find agent action execution calling the approval gate. Re-checked against engine edition 5f8ea3af: the gate is wired on every agent run path and covered by tests in the engine repository, so the entry is now beta.
- Knowledge ingestion pipelinesroadmapchanged tobeta
Fourteen connector inlets exist, but governed chunking, embeddings, lineage and ingestion orchestration are missing.
- Retrieval SLO dashboardsbetachanged toroadmap
Generic per-route latency percentiles exist, but there is no retrieval-specific latency, recall or freshness dashboard.
Retrieval
- BM25 full-text retrievalavailable
crates/core/src/personalities/search.rs— BM25 inverted index, Unicode tokenisation, ASCII folding, Rayon-parallel scoring. API:POST /v1/search. SDKs:search.ts,search.py.- Cosine vector retrievalavailable
crates/core/src/personalities/vector.rs+vector_hnsw.rs— brute-force cosine scan + HNSW graph. API:POST /v1/vector. Benchmark:2026-07-12-vector-scale.json.- Hybrid product retrievalavailable
crates/core/src/personalities/hybrid.rs— BM25 first-pass + facet filtering + optional cosine rerank. API:POST /v1/product-search.- Kind-scoped searchavailable
- Kind pre-filter on search and vector endpoints.
- Unicode-aware tokenisationavailable
- Unicode word segmentation + ASCII folding in
search.rs. - Incremental retrieval indexbeta
crates/core/src/personalities/vector_spann.rs— SPANN index structure.- Predicate-aware ANNbeta
rabitq.rs,caustic.rs,vector_hnsw.rs— filtered vector retrieval with recall@10 evidence.- Retrieval policy rerankingavailable
- No explicit ranking policy DSL.
- Multimodal retrievalresearch
- No implementation.
- Learned hybrid fusionresearch
- No implementation.
Models & embeddings
- Provider-agnostic completionsavailable
crates/llm/src/lib.rs—LlmProvidertrait,CompletionResponseshape.- Anthropic provideravailable
- Anthropic Messages API provider. Default model
claude-sonnet-4-5. - OpenAI-compatible provideravailable
- OpenAI-compatible
/v1/chat/completionsprovider. - Local Ollama provideravailable
- Local Ollama provider, no API key required.
- Provider fallback chainsavailable
LlmRegistry::complete_with_fallback— tries Ollama → Anthropic → OpenAI.- Embedding dimension contractsavailable
- Explicit embedding dimensions at write and query boundaries.
- Model routing policyroadmap
pricing.rsexists but no policy-based routing.- Managed embedding jobsbeta
- Implemented on branch
codex/t3-embedding-jobs(not yet gate-promoted):crates/embedding-jobs— managed embedding workload reusing knowledge-ingestion's durable-job primitives (JobState,RetryPolicy/backoff_delay,content_hashidempotency, project scoping) andkynetra_llm::EmbeddingProvider. Provider/model/dimension pinned per job; batched embed; idempotent re-runs (no re-embed of committed records, proven by a restart/CountingEmbedder test); bounded retry + per-record dead-letter queue; embeddings/sec rate limit; cancellation + checkpoints; stale-refresh and model migration (dimension change handled); project quotas; honest cost accounting (tokens/cost N/A, never fabricated). Additive/v1/embedding-jobs/*routes + OpenAPI. - Model cost observabilityroadmap
- Pricing data exists, no per-model cost dashboard.
- Fine-tuning registryresearch
- No implementation.
Agents & tools
- Structured tool definitionsavailable
Actiontrait with JSON Schemainput_schema(). 14 built-in actions.- Tool-call round tripsavailable
- Tool calls and results carried through message model.
MAX_TURNS = 8. - Memory collectionsavailable
crates/agents/src/memory/— craft, recall, consolidate, lore modules. API:POST /v1/memory/*.- Scoped operational toolsavailable
allowed_actionswhitelist on Agent entity.- Agent audit trailsavailable
AgentRunentities persisted in Universal Log. Queryable via filter DSL.- Approval gatesbeta
- Approval records, UI, timeout, and routes exist (
crates/agents/src/approvals.rs), but agent action execution never calls the approval gate. - Durable agent runsroadmap
- Scheduler exists but no checkpoint/resume for long runs.
- Agent regression suitesroadmap
- No implementation.
- Multi-agent role orchestrationresearch
orchestrate.rsexists but not production-ready.- Self-improving tool selectionresearch
- No implementation.
Data foundation
- Document entitiesavailable
crates/core/src/entity_store.rs— JSON entities as first-class log records. API: CRUD on/v1/entities.- Key-value accessavailable
crates/core/src/personalities/kv.rs— direct key reads.- Metadata filtersavailable
crates/core/src/personalities/doc.rs— JSON predicate filtering. API:POST /v1/filter.- Files alongside recordsavailable
crates/files/— file metadata and permissions as database records.- Events in the data planeavailable
crates/realtime/+crates/estuary/— events from mutation stream.- Typed schema viewsbeta
crates/schema/— typed views with policy enforcement.- Knowledge ingestion pipelinesbeta
crates/estuary/src/inlets/— 14 source connectors. Missing: governed chunking, embeddings, lineage, orchestration.- Document lineageroadmap
- No implementation.
- Semantic deduplicationresearch
- No implementation.
- Temporal knowledge queriesresearch
- No implementation.
Governance
- Auth and API keysavailable
crates/auth/— bcrypt + JWT, magic link, TOTP 2FA.- Roles and scopesavailable
crates/server/src/rbac.rs— project RBAC overlay.- Audit eventsavailable
crates/server/src/audit.rs— audit events in Universal Log.- Tenant boundariesavailable
crates/server/src/tenancy.rs— project-level isolation.- File permissionsavailable
- File access metadata aligned with application permissions.
- Retrieval policiesbeta
crates/schema/src/policy.rs— basic policy DSL.- PII redaction flowsroadmap
- No implementation.
- Data-residency policybeta
crates/compliance/src/residency.rs+crates/server/src/residency.rs— per-project, per-dimension placement policy: DENY-over-silent-placement decisions, provider validation, drift detection, and an HONEST per-dimension readiness (storage/compute/backup enforced; inference advisory; logs/metadata unenforced — never falsely green). Generalises the DPDP store indpdp.rs. Seedocs/capabilities/data-residency.md. Not a legal certification.- Policy simulationresearch
- No implementation.
- Provenance confidence scoringresearch
- No implementation.
Realtime context
- Change feedsavailable
crates/realtime/src/lib.rs— change feed from mutation stream.- SSE subscriptionsavailable
GET /v1/realtime— SSE event stream.- Realtime eventsavailable
- Realtime events from data-plane mutations.
- Durable conversation contextavailable
- Agent memory persisted as Universal Log entities.
- Local-first contextavailable
- Single binary with local WAL.
- Streaming retrieval updatesbeta
crates/agents/src/streaming.rs— streaming agent runs.- Live agent handoffroadmap
- No implementation.
- Event-driven agentsroadmap
- Cron scheduler exists (
crates/agents/src/scheduler.rs) but is not durable data-event consumption. - Causal context graphresearch
- No implementation.
- Collective memory controlsresearch
- No implementation.
Portability
- Open data planeavailable
- Rust engine, no cloud provider dependency.
- Single-binary deploymentavailable
kynetra-serverstandalone process.- Local and Docker workflowsavailable
Dockerfile+docker-compose.yml.- Cloud runtime profilesavailable
RuntimeProfileenum: dev-single, tiny-vps, prod-single, pg-ha-read, edge-read, etc.- Provider boundariesavailable
crates/providers/— provider-specific behavior outside storage kernel.- Edge-read profilebeta
- D1 bridge deployed at
edge.db.kynetra.dev. Not production: no automated Worker tests, vector scans capped at 5,000 candidates. - Portable backup bundlesroadmap
- No export/import bundles.
- Cross-cloud migrationroadmap
- No implementation.
- Air-gapped AI deploymentsresearch
- No implementation.
- Sovereign control planeresearch
- No implementation.
Operations
- Health contractsavailable
GET /healthz,GET /readyz— unauthenticated liveness/readiness.- Metrics and logsavailable
crates/server/src/db_metrics.rs+crates/observability/— per-route SLO tracker, slow query detection.- Admin operationsavailable
crates/server/src/db_admin.rs+infra_admin.rs— collections, keys, users, backups. Admin V2 SPA.- Backup-aware architectureavailable
- Checkpoint-based backup/restore architecture.
- Published benchmark evidenceavailable
benchmarks/run.sh+benchmarks/evidence/— documented benchmark context with methodology.- Retrieval SLO dashboardsroadmap
- Generic per-route p50/p95/p99 tracking exists. No retrieval-specific latency/recall/freshness dashboard.
- Model incident responseroadmap
- Implemented on branch
codex/t4-model-incident-response(not yet gate-promoted): pure domain incrates/schema/src/model_incident.rs(incident entity + enforced state machinedetected → acknowledged → mitigating → resolved \| closed, a CLOSED set of three bounded reversible mitigations — force-open breaker / add toblocked_models/ advisoryforce_exclude— a content-free append-only timeline, project-scoped durable store, and a pure error-class-spike detector reusing model-observability counts); orchestration + HTTP incrates/server/src/model_incident_api.rs(theIncidentEnginemaps each mitigation onto the *already-tested* model-routing levers behind agent-evals'ApprovalDecidergate — an IMPACTFUL mitigation cannot execute without an approval, so no destructive autonomous action occurs by default; resolving LIFTS every override; ephemeral overrides are re-driven at boot; a pluggable notification adapter defaults to log-only with no external send). Detection inherits model-observability's HTTP-route-only coverage caveat. Additive/v1/model-incident-response/*routes + OpenAPI. - AI capacity planningroadmap
- No implementation.
- Autonomous remediationresearch
- No implementation.
- Counterfactual retrieval debuggingresearch
- No implementation.
Developer surface
- HTTP APIavailable
crates/server/src/main.rs+rest.rs— full REST API.- TypeScript clientavailable
sdks/typescript/src/— 10 modules (auth, collections, entities, files, http, products, realtime, search, types).- MCP surfaceavailable
crates/mcp/— read-only MCP JSON-RPC surface at/mcp.- CLI lifecycleavailable
crates/cli/— init, start, status, install commands.- Rust coreavailable
crates/core/— native Rust engine.- SDK retrieval helpersbeta
sdks/typescript/src/search.ts+sdks/python/kynetra/search.py— basic helpers.- Prompt versioningroadmap
- No implementation.
- Local evaluation harnessavailable
- No implementation.
- Semantic debuggerresearch
- No implementation.
- Intent test generationresearch
- No implementation.
Evaluation & quality
- Measured retrieval benchmarksavailable
benchmarks/evidence/*.json— documented benchmark context with commit, hardware, methodology.- Retrieval benchmark workloadsavailable
crates/core/benches/+crates/core/examples/— BM25, vector, filtered vector workloads.- Honest capability statesavailable
ai-features.tsenforces evidence links foravailablefeatures.- Architecture quality gatesavailable
docs/scale/architecture-quality-gates.md— 64 quality gates.- Source-linked feature claimsavailable
- Evidence links in
ai-features.tspoint to source documentation. - Recall measurementbeta
- Reproducible benchmark code, brute-force ground truth, and evidence artifacts exist. Not publicly reachable without private-repository access.
- Labelled relevance datasetsavailable
- No implementation.
- Prompt regression trackingroadmap
- No implementation.
- Human preference evaluationresearch
- No implementation.
- Continuous trust modelresearch
- No implementation.