Glossary
Entries are representative examples (categories matter more than endorsing a single vendor). Tags below each entry use the same tag styling as the Tags lists in the AI Capabilities Stack diagram sidebar (monospace, surface background, plane-colored text and border): agentic stack planes for products and protocols, enterprise planes for org and strategy themes. Click a tag to show only glossary entries that include that plane (?plane= in the URL).
A2A
Agent-to-agent protocol for interoperable agents (e.g. Google A2A); often paired with agent cards for discovery.
A2A Agent Cards
Metadata for exposing and discovering agents in A2A-style ecosystems.
ACP
Agent Communication Protocol (related agent connectivity pattern; named in protocol-convergence copy as MCP to ACP to A2A).
AI fluency
Operating model and upskilling so teams can use AI responsibly in production, not only in pilots.
ANP
Agent networking / protocol family referenced alongside MCP and ACP in integrations-plane tooling.
Astrocytes (framework)
Open-source memory framework (retain / recall / reflect) between agents and storage: hybrid retrieval, per-bank access control, policy layer, pluggable Tier 1 retrieval backends and Tier 2 memory engines. Not an LLM gateway or agent runtime—orchestration stays in the harness; Astrocytes is the governed Context & Memory-style boundary. Repository: AstrocyteAI/astrocytes.
Arize AX
ML / LLM observability and evaluation platform (often used for monitoring and experiment tracking).
AWS Bedrock AgentCore
AWS managed runtime surface for agent execution (vendor-specific control/runtime path).
AWS Bedrock Guardrails
Managed policy and content safety controls for Bedrock-hosted models and flows.
Braintrust
Evaluation and logging platform for LLM applications (scores, datasets, experiments).
Budget gates
Spend and approval checkpoints (tokens, infra, vendors) tied to value milestones and portfolio decisions.
Build vs buy
Strategic choice between internal build, vendor product, or partner delivery for a capability or AI surface.
Cedar
Policy-as-code language (AWS Cedar) for fine-grained authorization and guardrails-style decisions.
Celery
Distributed task queue for Python (async and background workers).
Change management
Structured adoption and rollout: communications, training, champions, and feedback loops so new agent workflows stick.
Cloud Run
Google Cloud managed container execution (request-driven and worker-style workloads).
Cloudflare Workers
Edge serverless functions for routing, auth edges, or lightweight BFFs in front of agents.
CoreWeave
Cloud provider often used for GPU-heavy and ML inference workloads.
Datadog LLM Obs
Datadog product surface for LLM and GenAI observability (traces, metrics, usage).
Daytona
Sandboxed dev / agent runtime environments (isolated execution for coding agents).
Doppler
Secrets and config management for apps and CI (API keys, env sync).
E2B
Sandboxed code execution environments for agents (secure runtimes).
Executive sponsor
Accountable leader who owns funding, prioritization, and cross-functional unblock for an AI initiative.
Elasticsearch
Search and analytics engine; optional backing store for logs or hybrid retrieval.
Exa
Neural / semantic web search API for retrieval and research-style agents.
FastAPI
Python web framework for building MCP servers, tool APIs, and agent backends.
Fly.io Sprites
Fly.io execution model for lightweight edge/worker style processes (brand naming may evolve).
HashiCorp Vault
Secrets management, dynamic credentials, and encryption as a service.
Hindsight
Memory platform for long-horizon agents; uses TEMPR (Temporal Entity Memory Priming Retrieval) for structured retrieval over World, Experience, Opinion, and Observation networks (see TEMPR).
Inngest
Event-driven durable workflows and step functions in application code (retries, scheduling).
Intercom
Customer messaging and support surfaces where agent handoffs and approvals may appear.
Lambda
AWS Lambda (serverless functions) for sync APIs, workers, or tool endpoints.
LaunchDarkly
Feature flags and progressive delivery for safe rollouts (including model and policy toggles).
Langfuse
Open-source LLM engineering platform (tracing, evals, prompt management).
LangGraph
LangChain library for building stateful, cyclic agent workflows (graphs as control flow).
LangSmith
LangChain observability and evaluation suite (traces, datasets, testing).
LiteLLM
Unified API gateway to many model providers (routing, keys, fallbacks).
MCP (10K+ servers)
Model Context Protocol ecosystem (Anthropic-led); large public server corpus for tools and resources.
Mem0
Memory layer for AI apps (preferences, facts, retrieval APIs).
Modal
Python-first serverless compute for batch and GPU jobs (functions and sandboxes).
mTLS
Mutual TLS for workload-to-workload authentication (pairs with SPIFFE-style identity).
OAuth 2.1
Authorization framework baseline for user and delegated access to APIs and agents.
OPA
Open Policy Agent (Rego policies) for admission control, API policy, and guardrails integration.
OneCLI
CLI tooling category for unified secrets or cloud login flows (vendor-specific; name as used in diagram).
OpenAI Agents SDK
SDK for building agents with OpenAI models (tool calls, guardrails hooks).
OpenTelemetry
CNCF standard for traces, metrics, and logs (including GenAI semantic conventions for LLM spans).
Pinecone
Managed vector database for embeddings and similarity search.
pgvector
PostgreSQL extension for vector similarity (co-locate embeddings with relational data).
Playwright
Browser automation for scraping, testing, and tool-using agents against web UIs.
Portkey
AI gateway for models (routing, fallbacks, guardrails, observability hooks).
PostgreSQL
Relational database for tenants, behavior schemas, and operational data.
Psychological safety
Team norms where people can surface risks, failures, and questions without fear of blame - essential for safe AI experimentation.
Redis
In-memory data store for queues, caches, sessions, and rate limits.
Restate
Durable execution / virtual actors for long-running reliable workflows.
RFC 8693
OAuth 2.0 Token Exchange (swap tokens for different audience or type at the authorization server).
Risk appetite
Explicit bounds on what risk types and residual exposure the org accepts for AI (models, data, third parties, reputational).
RunPod
GPU cloud for inference and training workloads.
S3
Object storage (e.g. AWS S3) for artifacts, datasets, and large blobs.
Schema migrations
Versioned database schema changes (e.g. Flyway, Liquibase, Prisma migrate) for behavior and tool I/O contracts.
Sentry
Error tracking and performance monitoring for apps and agent pipelines.
Slack
Team chat surfaces where agents surface approvals, notifications, and human-in-the-loop flows.
SPIFFE
Secure Production Identity Framework for workload identity (SVIDs, trust domains).
Stripe
Billing and payments for productized agent features and metering.
Supabase
Postgres-backed BaaS (auth, storage, APIs) for app and agent backends.
Tavily
Search API oriented to LLM and agent retrieval use cases.
Teams
Microsoft Teams as a channel for enterprise agent and approval workflows.
Temporal
Durable workflow orchestration (long-running, retriable business processes).
TEMPR
Temporal Entity Memory Priming Retrieval — structured retrieval used by Hindsight across the four memory networks (World, Experience, Opinion, Observation).
Twilio
SMS, voice, and messaging APIs for notifications and human handoff channels.
Use-case portfolio
Prioritized backlog of AI opportunities with sequencing, dependencies, and kill criteria (often linked to build vs buy).
Unit economics
Per-task or per-outcome cost and revenue model (tokens, infra, support) for AI features and portfolios.
Vectorize.io
Vector / retrieval infrastructure vendor (embedding pipelines and search).
Vercel
Hosting and edge functions for web apps and agent frontends.
Zep
Long-term memory and conversation graph for agents.
Previous: Our methodology | Next: Learning tracks