Agentkeeper solved the Goldfish Memory problem in AI Agents.v1.1 out now

AgentKeeper released version 1.1 of its AI agent infrastructure, solving the "Goldfish Memory" problem by providing cognitive continuity across model switches, crashes, and restarts. The update introduces protected principles and constraints that survive compression cycles, TTL-based fact expiration for compliance, and an MCP server enabling any MCP-aware client to access the agent's cognitive layer without integration code.

Cognitive continuity infrastructure for long-lived AI agents. Your agent survives model switches, crashes, context-window limits, and restarts — with the same identity, memory, and priorities it had before. Agents don't fail because they forget facts. They fail because they lose cognitive continuity — their state, priorities, and identity drift the moment the model changes, the context window fills, or the process restarts. AgentKeeper treats this as a systems problem, not a memory problem. pip install agentkeeper-ai Zero required dependencies. No external infrastructure. Storage defaults to local SQLite. pip install 'agentkeeper-ai anthropic ' Claude pip install 'agentkeeper-ai openai ' GPT + OpenAI embeddings pip install 'agentkeeper-ai gemini ' Gemini pip install 'agentkeeper-ai semantic ' Local embeddings sentence-transformers pip install 'agentkeeper-ai mcp ' MCP server Claude Desktop, Cursor, Codex pip install 'agentkeeper-ai encrypted ' Encrypted storage at rest pip install 'agentkeeper-ai all ' Everything Principles and constraints are protected — exempt from every compression pass, injected into every reconstructed context, regardless of token budget. They survive decay, consolidation, contradiction arbitration, model switches, and process restarts. python import agentkeeper agent = agentkeeper.create agent id="aria", provider="anthropic" agent.set identity name="Aria", role="EU insurance broker copilot", principles= "never share PII without explicit consent" , constraints= "EU data residency only" , agent.principle "always confirm budget changes in writing" agent.fact "client: Acme Corporation", importance=0.95 agent.event "contract signed", when="2026-05-15" agent.save 100 compression cycles later — identity intact. The cognitive state is reconstructed in the format each model expects. XML for Claude, labelled sections for GPT-4, narrative prose for Gemini, terse tokens for Ollama. One agent, four runtimes, zero rewrites. agent = agentkeeper.load "aria", provider="anthropic" response = agent.ask "What do we know about Acme?" agent.switch provider "openai" .save response = agent.ask "Same question, different model." Memory and identity are intact. Format has changed. Nothing broke. Facts and graph triples accept a TTL. When it lapses, purge expired removes them. No manual cleanup. Compliant by default. agent.fact "session token: abc123", ttl="1h" agent.fact "audit log reference: Q1-2026", ttl="90d" agent.link "Acme", "signed contract", "ThinkLanceAI", ttl="2y" agent.purge expired removes what's lapsed, keeps what's protected Facts are prose. Triples are structure. Both live in the same agent, with their own retention and TTL policy. agent.link "Acme", "owns", "Globex" agent.link "Globex", "located in", "BE" agent.link "Alice", "works at", "Acme", confidence=0.9 related = agent.find related "Acme", max hops=2, direction="out" {"Globex": 1, "BE": 2, "Alice": 1} AgentKeeper ships an MCP server. Any MCP-aware client — Claude Desktop, Cursor, Claude Code — gets full access to the agent's cognitive layer without writing a line of integration code. agentkeeper-mcp --agent-id aria --provider anthropic claude desktop config.json : { "mcpServers": { "aria": { "command": "agentkeeper-mcp", "args": "--agent-id", "aria", "--provider", "anthropic" } } } Available tools over MCP: add fact , recall , set identity , link , find related , compress , health , gdpr export , purge expired , checkpoint , restore , list checkpoints . Freeze the full cognitive state into an immutable, content-hashed snapshot. Restore it after a crash, a context-window overflow, a model switch, or a process restart. Attach an opaque execution state payload current file, pending task, todos — AgentKeeper stores and returns it verbatim; it never interprets or runs it. snap = agent.checkpoint label="before refactor", execution state={ "current file": "auth.py", "pending task": "finish RS256 migration", }, crash, restart, switch model — the process memory is gone agent = agentkeeper.load "aria" .restore snap.snapshot id identity, facts, graph, and execution state are back agentkeeper.diff snap a, snap b factual diff: facts added/removed/modified agent.list checkpoints every snapshot, oldest first Reconstruction is deterministic — the same snapshot always rebuilds the same cognitive state, verified by a content hash. Behaviour is not guaranteed: AgentKeeper restores the agent's state, not the model's next decision. The crash-recovery demo runs with no API key: python examples/crash recovery.py ┌──────────────────────────────────────────────────────────────┐ │ AgentKeeper Public API │ │ agent.remember · agent.recall · agent.ask │ │ agent.compress · agent.link · agent.find related │ │ agent.set identity · agent.purge expired · agent.save │ └────────────────────────────┬─────────────────────────────────┘ │ ┌────────────────────────────▼─────────────────────────────────┐ │ Cognitive Reconstruction Engine CRE │ │ Identity injection · importance ranking · semantic boost │ │ Token budget · profile-driven rendering │ └─┬───────────┬────────────┬────────────┬──────────────────────┘ │ │ │ │ ┌─────▼─────┐ ┌──▼─────────┐ ┌▼──────────┐ ┌▼──────────────┐ │ Memory │ │ Semantic │ │ Cognitive │ │ Cross-Model │ │ Hierarchy │ │ Recall │ │ Compress │ │ Translation │ │ │ │ │ │ │ │ │ │ working │ │ embeddings │ │ decay │ │ XML Claude │ │ episodic │ │ vector idx │ │ consol. │ │ sections GPT │ │ semantic │ │ sqlite-vec │ │ contradic │ │ narrative G. │ │ archival │ │ │ │ │ │ minimal Oll. │ └───────────┘ └────────────┘ └───────────┘ └───────────────┘ │ │ ┌─────▼────────────────────────────────────────────▼────────────┐ │ Graph Layer Storage pluggable │ │ Triple, TTL, BFS SQLite · Encrypted SQLite · Postgres │ │ agent.link AGENTKEEPER DB, AGENTKEEPER ENC KEY │ └───────────────────────────────────────────────────────────────┘ │ ┌─────────────────────▼───────────────────┐ │ MCP Server · LangChain · CrewAI │ │ agentkeeper-mcp · langchain system prompt│ └──────────────────────────────────────────┘ Postgres stub available; full implementation in v1.2. agent.fact "budget: 50k EUR", importance=0.9 stable semantic fact agent.event "contract signed", when="2026-05-15" episodic, time-anchored agent.principle "never share PII" protected, survives all compression agent.remember "favourite colour: blue" tier inferred automatically results = agent.recall "money allocated to the project", top k=5 for fact, score in results: print f"{score:.2f} {fact.content}" Pluggable backends: local sentence-transformers default, free, offline , OpenAI, or your own. Persistent index via sqlite-vec — survives process restarts without rebuild. report = agent.compress CompressionReport decayed facts=12, consolidation={'clusters found': 3, 'facts removed': 7}, contradictions={'pairs found': 2, 'resolutions': 2}, facts before=120, facts after=102, Three independent passes: decay exponential half-life on unused facts , consolidation embedding-based clustering, optional LLM synthesiser , contradiction arbitration key-value divergence + polarity detection, deterministic winner . Protected facts are immortal. agent.link "Acme", "owns", "Globex", confidence=1.0 agent.link "Acme", "signed contract", "ThinkLanceAI", ttl="2y" BFS traversal related = agent.find related "Acme", max hops=2, direction="out" Introspect triples for triple in agent.triples: print triple Triple 'Acme' - owns - 'Globex', conf=1.00 agent.fact "session token: abc123", ttl="1h" agent.fact "temp context: negotiation", ttl="30d" agent.link "Project Phoenix", "uses provider", "Anthropic", ttl="P90D" agent.purge expired returns count of removed items python from agentkeeper.storage.encrypted sqlite import EncryptedSQLiteStorage key = EncryptedSQLiteStorage.generate key storage = EncryptedSQLiteStorage encryption key=key or via env: AGENTKEEPER ENCRYPTION KEY=... AES-128-CBC + HMAC-SHA256 at rest. Progressive migration — plain SQLite rows are upgraded on next save. python import asyncio, agentkeeper async def main : agent = agentkeeper.create async agent id="aria", provider="anthropic" agent.set identity name="Aria", role="copilot" agent.fact "budget: 50k EUR", importance=0.95 answers = await asyncio.gather agent.ask "status?", provider="anthropic" , agent.ask "status?", provider="openai" , asyncio.run main Sync and async agents share the same storage. Save with one, load with the other. python from agentkeeper.integrations.langchain import LangChainCognitiveProvider from langchain core.prompts import ChatPromptTemplate provider = LangChainCognitiveProvider agent, model="gpt-4o" prompt = ChatPromptTemplate.from messages "system", provider message="What's our budget?" , "human", "{input}" , python from agentkeeper import CognitiveProfile, PromptFormat, register profile register profile CognitiveProfile provider="my-llm", format=PromptFormat.SECTIONS, effective context tokens=10 000, export = agent.gdpr export all facts, triples, identity — JSON-serialisable agent.purge expired remove anything whose TTL has elapsed agent.forget fact id remove a single fact by id agentkeeper.delete "aria" permanently remove an agent from storage snapshot = agent.health { "total facts": 42, "critical facts": 3, "protected facts": 5, "contradicted facts": 1, "stale facts": 4, "importance stats": {"mean": 0.71, "min": 0.10, "max": 0.95}, "tier distribution": {"semantic": 28, "episodic": 10, ...}, "graph": {"total triples": 8, "protected triples": 2}, "identity": {"name": "Aria", "principles count": 3, ...} } Type-safe — py.typed shipped, mypy-strict compatible. Typed exceptions — AgentKeeperError root, subclasses for every failure mode. Structured logging — namespaced under agentkeeper. , opt-in, NullHandler default. Retries — exponential backoff + jitter via with retry / with async retry . Tested — 491 tests, CI on Python 3.10 / 3.11 / 3.12. Zero breaking changes from v0.1 — agent.remember content, critical=True still works. | Variable | Default | Purpose | |---|---|---| OPENAI API KEY | — | OpenAI provider | ANTHROPIC API KEY | — | Anthropic provider | GEMINI API KEY | — | Gemini provider | OLLAMA HOST | http://localhost:11434 | Ollama server | AGENTKEEPER DB | agentkeeper.db | SQLite path | AGENTKEEPER ENCRYPTION KEY | — | Fernet key for encrypted storage | AGENTKEEPER EMBEDDING PROVIDER | auto | sentence-transformers › openai › mock | pip install agentkeeper-ai python examples/demo.py The demo runs entirely on provider="mock" and AGENTKEEPER EMBEDDING PROVIDER=mock . No keys. No network. Shows identity hardening, compression, cross-model translation, and async — in one file. v1.1 ✅ — TTL, graph layer, encrypted storage, MCP server, LangChain + CrewAI integrations, persistent sqlite-vec index, GDPR export, health snapshot, cognitive checkpoints snapshot / restore / diff , 491 tests. v1.2 — Postgres storage backend full implementation , TypeScript SDK. v1.3 — AgentKeeper Cloud managed sync . OSS stays feature-complete. Issues, ideas, and PRs welcome. See CONTRIBUTING.md /Thinklanceai/agentkeeper/blob/main/CONTRIBUTING.md . MIT. See LICENSE /Thinklanceai/agentkeeper/blob/main/LICENSE . ThinkLanceAI https://thinklanceai.com — Tom Anciaux Berner — cognitive infrastructure for AI systems. tom@thinklanceai.com