Glossary · A–Z

The Agent Vocabulary.

Every term your coding agent assumes you know, defined, with the Anvaya implementation attached.

A–Z

Every Term, Sourced.

01Context RotWhy more context means worse answers.02CompactionSummarize the past to afford the future.03Prompt CachingPay full price once for stable context.04EVOI Proactive InjectionMemories that volunteer before they're asked.05Claim Keys & AdjudicationKnowledge graduates on evidence, not optimism.06PackedContext2–4K tokens that beat 50K raw.07Token-FoldCompression in three tiers.08Beta-Bernoulli UtilityScore memories by whether they helped.09Just-in-Time RetrievalLook it up the moment it matters.10Per-Type Memory Decay (Gravitas)Forgetting on a schedule that matches reality.11Drift Detection (Git-Anchored Invalidation)The codebase is the ground truth.12AGENTS.mdOne instruction file every agent reads.13CLAUDE.mdClaude Code's standing orders.14MCP Memory ServerOne memory, every harness.15Subagents & IsolationFresh windows for dirty work.16Headless Coding Agents (CI Mode)Agents that run where no terminal watches.17Plan Mode (Read-Only Exploration)Look before the agent leaps.18Repomap (Code Graph Retrieval)Who calls this, answered in tokens, not files.19Shadow EvaluationCanary deploys for knowledge.20AI Coding Token Costs (2026)Price the input, not the prose.21Agentic SwarmMany agents, one ledger, one gate.22Worktree PlacementGit-native isolation without the clone tax.23Footprint vs RSSTwo memory numbers, one truth on macOS.24Sandbox PolicyApproval and confinement are different axes.

Questions

About This Glossary.

Q

Why an agent glossary?

The field converged on vocabulary fast (context engineering, EVOI, claim keys, shadow eval) but definitions scatter across papers and blog posts. One page per term, each with how Anvaya implements it, gives humans and AI search a canonical reference.

Q

Are these Anvaya-specific terms?

No, most are industry terms (Anthropic, LangChain, MemCoder literature) with an Anvaya implementation note. Proprietary mechanisms (Token-Fold, PackedContext, Gravitas) are marked as Anvaya's.

Run Agents That Fit On Your Laptop.

25.6 MB median RSS. 25 agents ran in parallel on a Core 2 Duo with 4 GB RAM. Hundreds on your machine. Zero cloud required on the Ollama path.

Get Started →Read the FAQ

Requires Rust/cargo to build from source. Linux and macOS today, Windows not yet supported. Pre-1.0, public beta. Pricing TBD.