Open infrastructure for persistent AI cognition

The past should change the future.

AI agents should not start from zero every time. Anda builds the open protocols, infrastructure, and benchmarks that let agents preserve cognitive state, learn from experience, and carry what they learn into future decisions.

Explore KIP 2.0 Benchmark with MIB

Memory Experience Skills Provenance Governance Evaluation

Anda.ai stewards the protocols, infrastructure, and benchmarks. Products such as Anda Brain and Anda Bot are built on top by Yiwen.AI.

The problem

Storage is not memory.

Logs preserve the past.

Retrieval finds the past.

Memory lets the past participate in the future.

A long context window postpones forgetting. A vector database returns related text. Neither explains what the agent was trying to achieve, which action changed the situation, where reality contradicted expectation, or when a lesson should - and should not - affect the next decision.

The cross-temporal cognitive loop

Past
Experience
Memory formation
Cognitive state
Recall & revision
Future decision
Action
Outcome

Every outcome becomes the next experience. What is being evaluated is not a retriever - it is the agent and its memory as one cross-temporal cognitive system.

KIP

defines durable cognition.

Anda

builds it.

MIB

measures whether it works.

KIP 2.0 · Cognitive State Protocol 2.0 Draft

A memory brain needs more than facts.

KIP 2.0 is an open protocol for persistent cognitive state. It gives AI agents a structured way to represent not only what can be said, but what they believe, why they believe it, what happened to them, what they learned, and who has authority over that cognition.

Meaning ≠ Belief ≠ Authority A proposition is truth-neutral. Only an assertion carries epistemic commitment. Provenance is not authority.

Semantic state

Represent what can be said.

Concepts, propositions, schema packages and structured relations - including statements about statements.

Epistemic state

Represent what is believed - and why.

Assertions, evidence, confidence, contradiction and provenance. Conflicting assertions coexist instead of overwriting each other.

Temporal state

Remember change, not just the latest value.

Valid time, observation time and assertion time stay independent, so current truth never erases historical truth.

Experience & skill

Preserve what happened. Learn what tends to work.

Experience captures goal → action → observation → feedback → outcome. Skill compiles reusable experience into procedural cognition.

Governance

Cognition has owners and boundaries.

Memory Spaces define ownership, isolation, access and mutation authority at the Cognitive Nexus - not in a prompt.

Portable cognition

Move memory without flattening its meaning.

Transactions, Capsules, capability negotiation, schema packages and a conformance suite make cognitive state interoperable.

One proposition, two actors, no corruption

Proposition truth-neutral

(:alice, "prefers", :dark_mode)

Assertion · support

by: :alice
mode: "stated"
confidence: 0.95
evidence: :msg_2f1

Assertion · reject

by: :bob
mode: "stated"
confidence: 0.60
evidence: :msg_9c4

Belief projection contested

Disagreement is representable state, not corruption. Belief is projected at read time from assertions and evidence - it is never stored as truth.

KML · form and revise cognition

ASSERT (:alice, "prefers", :dark_mode) {
  by: :alice,          // the semantic actor
  mode: "stated",      // observed|stated|inferred|...
  confidence: 0.95,    // strength of THIS stance
  evidence: :msg_2f1
}

// Bob disagrees. Both assertions coexist -
// contradiction is state, never corruption.
ASSERT (:alice, "prefers", :dark_mode) {
  by: :bob, mode: "stated", stance: "reject"
}

Read the KIP 2.0 specification KIP on GitHub

Experience → Knowledge → Skill

From experience to skill.

Knowledge captures reusable regularities. Experience preserves the path through goals, actions, observations and outcomes. Skill turns successful experience into reusable procedure.

Experience

What happened.

A bounded, goal-directed trajectory - not a chat transcript. The structure of the attempt, including the failure and the recovery, is itself a first-class cognitive object.

Goal
Action
Observation
Feedback
Outcome

↓ consolidation

Knowledge

What tends to be true.

Consolidation compresses repeated evidence and experience into stable regularities that still point back to what produced them.

↓ compilation

Skill

What tends to work.

Compilation turns successful experience into an action-selecting policy - including the applicability boundary where it should not be used.

Agents should remember not only what they know, but what they went through.

Memory Intelligence Benchmark Draft

Remembering is easy to claim. Hard to prove.

MIB is an open benchmark for measuring whether memory actually improves future cognition and behavior - not how much of the past a system can retrieve.

Did the right part of the past change the future in the right way?

KIP

What can durable cognition represent?

Anda

How can we build it?

MIB

Does memory actually make the agent better?

Traditional memory benchmarks

Store
Retrieve
Answer

MIB

Experience
Formation
Memory
Revision
Recall
Decision
Behavior
Outcome

Paired intervention

Same future task

Full memory

The agent keeps everything it formed.

Outcome A

Relevant memory ablated

Only the memory that should matter is removed.

Outcome B

The gap between A and B is the causal contribution of memory.

A memory is only useful if removing it changes what happens next.

Memory Benefit

How much better is the outcome when the relevant memory is present?

Memory Harm

How much worse does stale or incorrect memory make the system?

Irrelevant Memory Stability

Does unrelated history quietly steer future behavior?

MIB Core - six implemented dimensions

The main MIB Score measures absolute memory-enabled capability. Causal metrics are reported alongside it, never folded into one opaque number.

Retention & Retrieval

Can relevant past information be recovered accurately, including under indirect cues and interference?

Temporal Memory

Can the system separate current state, historical state, transitions, revisions and stale information?

Epistemic Memory

Can it remember who said what, preserve uncertainty, handle correction and contradiction, and not treat missing evidence as false?

Experience Memory

Can it preserve the structure of goals, actions, observations, failures, recovery and outcomes?

Skill Learning & Transfer

Can repeated experience become reusable policy - including knowing when not to transfer a learned skill?

Causal Memory Impact

Can we show that relevant memory improves future behavior while irrelevant, stale or harmful memory does not control it?

More profiles coming: Selective ForgettingProspective MemorySelf MemoryCross-Agent MemoryMultimodal Memory

You don't need KIP to run MIB.

MIB is deliberately architecture-neutral. Vector memory, summary memory, graph memory, episodic memory, procedural memory, full-context replay, KIP memory or any hybrid can participate on equal terms.

Explore MIB

How it fits together

One stack, measured end to end.

From a sovereign foundation up to the agents you can use today - with an evaluation plane that keeps the whole system honest.

Evaluation plane

MIB

Memory intelligence measured across the whole system rather than at any single layer. MIB is not a runtime tier - it is an instrument laid across the agent and its memory.

Applications

Agents in the world

Anda BotThird-party agents

Cognitive intelligence

Anda Brain

FormationRecallMaintenance

Cognitive infrastructure

Engine & runtime

Anda DBAnda runtime

Open protocols

Shared languages

KIP 2.0Agent Protocols

Sovereign foundation

Decentralized & verifiable

Ed25519 identityTEEInternet Computer

Anda DB

Persistent cognitive state engine

A high-performance Rust engine for KIP-native cognitive state, combining structured semantics, graph relations, vector retrieval and full-text search.

Graph + vector + BM25 + HNSW
KIP-native cognitive nexus
Multimodal data
Pluggable, encrypted storage

github.com/ldclabs/anda-db

Anda

Agent runtime

A composable Rust framework for building autonomous, multi-agent systems with cryptographic identity and confidential execution.

On-chain ICP identity
Composable tools & skills
TEE-ready, privacy first
Multi-agent orchestration

github.com/ldclabs/anda

Two axes of an agent

An Internet of Sovereign Minds

KIP gives an agent continuity with its past. Agent Protocols give it relationships with other agents. Together they describe a mind that persists and a network it can belong to.

KIP 2.0

Internal continuity

Cognitive state that survives the session: memory, experience, belief, evidence and the governance around them.

Cognitive stateMemory & experienceProvenance & governance

KIP on GitHub

Agent Protocols

External relations

An open specification suite so independent agents can identify, describe and talk to each other - without a central platform owning the conversation.

Agent IdentityAgent ProfileAgent Discourse

Agent Protocols on GitHub

Memory gives identity through time. Protocols give relationships across minds.

For developers

Start building in minutes

Everything is open source under permissive licenses. Run a Cognitive Nexus, ground your agent, then measure what its memory is actually worth.

Run a Cognitive Nexus

Spin up the KIP 2.0 server, or embed Anda DB directly through the Rust or Python bindings.

Ground before you write

One DESCRIBE PRIMER hands the agent its identity, Space, schema packages, capabilities and safety invariants. The Nexus describes itself.

Connect, then measure

Speak KIP over JSON-RPC, or front it with Anda Brain and any MCP client - then run MIB to see whether the memory earns its place.

KIP MIB Agent Protocols Anda Anda DB

quickstart.sh

# 1 - Run a Cognitive Nexus (KIP 2.0 over JSON-RPC)
cargo install anda_cognitive_nexus_server
anda_cognitive_nexus_server --db ./data

# 2 - Ground the agent before it writes
curl -sX POST http://localhost:8080/kip \
  -d '{ "command": "DESCRIBE PRIMER" }'

# 3 - Recall a belief, not just a row
curl -sX POST http://localhost:8080/kip \
  -d '{ "command": "FIND(?b.status) WHERE {
         ?b BELIEF (:alice, \"timezone\", ?tz) }" }'

Built on Anda

The stack, put to work

Applications engineered by Yiwen.AI on Anda’s open foundation.

The standalone "Anda AI" app is retired. Meet its successor, Anda Bot.

Anda Brain

Cognitive memory runtime

Turns raw interaction and experience into durable cognitive state - forming memories, recalling what matters, consolidating experience, and evolving skills over time.

FormationRecallConsolidationLearning

Learn more

★ Recommended

Anda Bot

Your agent on the open stack

An AI agent built on KIP and Anda Brain - it remembers you across sessions and acts on your behalf. The successor to the Anda AI app.

Try Anda Bot

Built in the open

An agent cognition commons

Anda is developed openly by LDC Labs and the ICPanda DAO. The protocols, the infrastructure and the benchmark all belong to everyone building on them.