01
Capabilities, not model dependencies
Chat, generation, summarization, extraction and classification behind versioned APIs and SDKs.
Intelligence as a platform
Connect models, knowledge, memory and agents through one governed platform. Give every application intelligence that understands its context.
The SoftKit AI ecosystem
Applications own business workflows. AI Foundation connects context, reasoning and execution. Infrastructure runs inference; Admin Console administers the platform.
Why AF
Unify model access, organizational knowledge, agent execution and policy. Reuse the same foundation across assistants, document workflows and software development.
Intelligence as a platform
01
Chat, generation, summarization, extraction and classification behind versioned APIs and SDKs.
02
Plan, act and observe with tools, step budgets, approval checkpoints and controlled autonomy.
03
Connect organizational sources, semantic retrieval and persistent memory to every conversation.
04
Assemble relevant context within token budgets. Version, render and evaluate reusable instructions.
05
Identity, tenant isolation, policies, protected secrets and input/output controls accompany every action.
06
Trace execution, attribute consumption, apply budgets and coordinate capacity across applications and replicas.
Portfolio
Twelve connected capability families. Open any family to see what it brings to your applications.
Chat, streaming, completion, summaries, extraction and classification; JSON Schema validation and bounded repair; native structured output, partial-result streaming and schema inference from Go types.
Source connectors, ingestion, chunking, embeddings, vector search, reranking, reindexing and retention. Hybrid ranking, semantic chunking, multimodal ingestion, provenance, ontology traversal and governed federated knowledge.
Session, agent, user and organization memory; write, search, compact, forget and export. Token budgeting, deterministic assembly, adaptive ranking, shared sessions, multimodal context and context caching. Federated memory, custom scopes and replay. Versioned prompt templates, variables and conditions, test-gated publication, no-code multilingual authoring, LLM-as-judge evaluation, feedback optimization and shared template marketplaces.
Plan–act–observe, permissions, approvals, cancellation and resumability. Multi-agent coordination, replanning, parallel tools, dynamic definitions, declarative pipelines, sandboxed code execution and agent/tool marketplaces.
Capability-based routing, stable aliases, retry, fallback and real-time streams. Self-hosted endpoints, model discovery, custom aliases, A/B and canary splits, adaptive selection, speculative execution and fine-tuning pipelines.
Verified identity, exact administrative scope, tenant isolation, PII redaction and output filters. Injection detection, quarantine, secret rotation, regional residency, DLP, customer-managed keys, moderation and behavioral anomaly detection.
Versioned policies, approvals, publication and activation. Policy simulation, decision history, configuration validation and rollback; integrity-verifiable audit export and federated security policies.
Source-recorded usage, workspace/application showback, CSV/JSON export and budget alerts. Cost-aware alias substitution, response caching, distributed quotas, leased reservations and FIFO provider admission.
Durable jobs and item results, concurrency limits, cancellation and recovery. Provider batch or Router dispatch, summarize/extract pipelines, cross-tenant packing, priorities, webhooks and submission reconciliation.
Traces, structured logs, metrics and health. Ratings, corrections and flags; opt-in content inspection, quality scoring, regression analysis, aggregate benchmarks, cost/latency anomalies and long-term telemetry retention.
General chat, software development, embeddings, retrieval and reranking; document analysis, OCR, image understanding, speech recognition and synthesis, image generation, video understanding, reasoning, planning and simulation.
Typed services, module lifecycle, health and controlled shutdown. Dynamic/remote modules, hot reload, marketplaces, SDKs, external APIs, self-service catalogs, multi-region and dedicated deployments, online migrations and adaptive scaling.
Stable contracts
Architecture
AI Services is the application boundary. Context, agents, memory and knowledge cooperate through explicit contracts. AI Router selects execution; connectors isolate providers. Kernel owns composition and lifecycle.
Governed execution
The outcome
“Connect your applications to shared intelligence, organizational knowledge and governed action.”
SOFTKIT AI FOUNDATION
Build assistants, document workflows and software automation on shared AI capabilities.
Back to the beginning