惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Microsoft Azure Blog
Microsoft Azure Blog
J
Java Code Geeks
量子位
腾讯CDC
C
Check Point Blog
小众软件
小众软件
IT之家
IT之家
I
InfoQ
Hugging Face - Blog
Hugging Face - Blog
Stack Overflow Blog
Stack Overflow Blog
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
GbyAI
GbyAI
Apple Machine Learning Research
Apple Machine Learning Research
大猫的无限游戏
大猫的无限游戏
博客园_首页
S
SegmentFault 最新的问题
The Cloudflare Blog
阮一峰的网络日志
阮一峰的网络日志
aimingoo的专栏
aimingoo的专栏
P
Proofpoint News Feed
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
Google DeepMind News
Google DeepMind News
T
Tailwind CSS Blog
Martin Fowler
Martin Fowler

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - canis-minor/typedmem: TypedMemory
ruxiz · 2026-05-17 · via Hacker News - Newest: "AI"

Long-term memory and reflection for AI agents. Persistent, evolving, context-aware — improves agent behavior over time.

CI PyPI Python License: MIT

📦 PyPI · 📚 Docs · 🏷️ Releases · 📝 Changelog

TL;DR

TypedMemory gives AI agents long-term memory.

  • remember new information
  • recall relevant context
  • reflect and improve over time

The problem

AI agents start believing their own hallucinations. They:

  • contradict themselves silently — the last write wins, the conflict disappears
  • overwrite past decisions with no audit trail — you can't debug what you can't see
  • never resolve goals — yesterday's "I'll do X" looks identical to today's "I did X"

TypedMemory makes that visible.

The contradiction-detection moment

$ pip install typedmem

$ typedmem --profile engineering_design add \
    "SQLite handles our single-writer load fine" --type risk --subject storage
$ typedmem --profile engineering_design add \
    "SQLite blocks under concurrent writes"     --type risk --subject storage

$ typedmem --profile engineering_design contradictions

1 contradiction cluster(s):

cluster 1 (2 memories):
  [risk] [storage] SQLite handles our load fine
  [risk] [storage] SQLite blocks under concurrent writes

Two memories cross-linked by the FLAG policy. Both still in the store — no silent overwrite. Run typedmem history <id> on either to see exactly when and why the state changed.

5 lines for an agent

from typedmem import AgentMemory

mem = AgentMemory(profile="personal", path="agent.db")

mem.remember("User wants to learn Rust by year end")
mem.remember("User lives in Tokyo")

hits = mem.recall("what is the user trying to learn?")
#   → [ScoredMemory(content="User wants to learn Rust...", score=0.78)]

report = mem.reflect()
#   → AgentMemoryReflection(contradictions=[], drift_records=[], ...)

Four verbs over the whole pipeline: remember (extract + store), recall (semantic retrieval), reflect (run the evolver pipeline), forget (explicit delete).

More demos: examples/DEMO.md for the 30-second no-flags paste · examples/agent_loop_demo.py for the before-vs-after agent story.

Before vs After

Without TypedMemory With TypedMemory
Agent changes its mind Last write silently overwrites REPLACE policy logs every change to replace_log; PreferenceDriftDetector flags instability
Two facts contradict One overwrites the other; you'll never know FLAG cross-links both; typedmem contradictions surfaces the cluster
A decision gets revised Old decision lost SUPERSEDE keeps the audit trail (old.superseded_by → new.id); typedmem history shows the lifecycle
Goals accumulate They sit there forever, mixing with current intent GoalResolver matches incoming events to active goals and flips them to resolved
Same fact arrives from 3 sources 3 duplicate memories REINFORCE merges into one, unions sources by (document_id, chunk_id, span), boosts confidence
Stale events pile up Search noise grows SummaryEvolver condenses non-destructively; originals link forward via metadata["summarizes"]

What makes TypedMemory different

Most systems store memory. TypedMemory evolves it.

  • contradictions are surfaced (not overwritten)
  • memory changes are tracked over time (every conflict resolution leaves a record)
  • goals are resolved automatically when matching evidence arrives
  • stale memory is summarized non-destructively
  • every change has an audit trailtypedmem history <id>

Memory becomes a living knowledge layer, not a log.

Use cases

Primary:

  • Debugging hallucinating agents. When an agent flips its story, run typedmem history <id> and see every state change with reason, timestamp, and previous content. Contradictions surface via mem.reflect() instead of disappearing under the next write.
  • Long-term agent memory — preferences, goals, drift. mem.remember() captures each session's signal. mem.recall() lets the next session see the current state. mem.reflect() catches preferences that keep flipping and goals that match recent events.

Also good for:

  • Multi-document research / RAG with structured provenance — Source(document_id, chunk_id, span, authority) per memory; REINFORCE merges duplicates across papers
  • Design-doc agents — decisions SUPERSEDE rather than overwriting; full audit trail
  • Multi-tenant agents (legal + medical + customer-success on one machine) — workspace isolates each domain

How it works

                              ┌──────────────────┐
                              │  DomainProfile   │  ← schema: which types,
                              │  TypeSpec × N    │     which policies,
                              │  prompt + rules  │     which validations
                              └────────┬─────────┘
                                       │
       text ──► Extractor ──► Memory ──┴──► MemoryStore ──► Retriever
                                            │
                                            ▼
                                         Evolver
                              (contradictions, drift, goals,
                               non-destructive summarization)

Every memory has a type (claim, decision, observation, …), a confidence, a structured source, a lifecycle policy, and a workspace — not a string in a vector database. Memories know how to update themselves on conflict, how to decay over time, and how to be summarized.

Zero runtime dependencies. Stdlib only. LLM clients, YAML profile loading, and richer embedders are optional extras.

Why this exists

Most "AI memory" libraries are wrappers around a vector database. That works for "remember what the user said," but it falls apart the moment you want an agent to:

  • track who said what, in which document, at which span (provenance)
  • handle the same fact from three sources without storing it three times (reinforcement)
  • recognize that a new decision supersedes the old one without losing the audit trail
  • summarize stale events without throwing away the originals
  • isolate legal memory from medical memory on the same machine
  • flag contradictions instead of silently overwriting them

TypedMemory handles these as first-class concepts, not bolt-ons.

Install

pip install typedmem                       # default install, zero deps
pip install 'typedmem[anthropic]'          # + AnthropicClient
pip install 'typedmem[openai]'             # + OpenAIClient
pip install 'typedmem[yaml]'               # + DomainProfile.from_yaml()
pip install 'typedmem[all]'

Python 3.10+.

60-second demo: an engineering design agent

import json
from typedmem import (
    DomainProfile, FakeClient, LLMExtractor, SQLiteMemoryStore,
)

profile = DomainProfile.builtin("engineering_design")
store = SQLiteMemoryStore.for_profile(profile, "design.db")

# Pretend the LLM extracted these from your design docs.
extractor = LLMExtractor(client=FakeClient([
    json.dumps([
        {"type": "decision", "content": "Use SQLite for storage",
         "subject": "storage_backend", "confidence": 0.9,
         "source": {"document_id": "design_v1.md"}},
        {"type": "risk", "content": "SQLite is single-writer",
         "subject": "storage_backend", "confidence": 0.8,
         "source": {"document_id": "design_v1.md"}},
    ]),
    json.dumps([
        {"type": "decision", "content": "Switch to PostgreSQL for concurrent writes",
         "subject": "storage_backend", "confidence": 0.9,
         "source": {"document_id": "design_v2.md"}},
        {"type": "risk", "content": "Postgres adds an external service",
         "subject": "storage_backend", "confidence": 0.85,
         "source": {"document_id": "design_v2.md"}},
    ]),
]), profile=profile)

for snippet in ("v1 text", "v2 text"):
    for m in extractor.extract(snippet):
        store.add(m)

# decision → SUPERSEDE: old preserved, new active.
print(store.by_type("decision"))                       # → just PostgreSQL
print(store.by_type("decision", include_superseded=True))  # → both

# risk → FLAG: two risks on the same subject get cross-linked.
for cluster in store.contradictions():
    for m in cluster:
        print(m.content)                                # → both risks

See examples/engineering_design_demo.py for the full version with audit trail and source provenance, or run:

typedmem profiles
typedmem --profile engineering_design add "..." --document-id design_v3.md
typedmem --profile engineering_design list --type decision
typedmem evolve --evolver contradictions

The mental model

Layer What it gives you Examples
Memory Typed object with content + confidence + workspace + sources + status Memory(type="claim", content=..., sources=[Source(...)])
Source Structured provenance with hashable identity (document_id, chunk_id, span) — dedup key for REINFORCE
workspace Namespace on every memory One agent, multiple corpora, zero cross-contamination
ConflictPolicy What to do when a new memory hits the same (workspace, type, subject) slot REPLACE · KEEP_BOTH · SUPERSEDE · REINFORCE · FLAG · IGNORE
DomainProfile Schema for a domain: which types, what policy each obeys, what's required engineering_design · research_paper · legal · medical_literature · personal · …
Evolver Reads memories (not text); produces audit-trailed actions ContradictionSurfacer · PreferenceDriftDetector · GoalResolver · SummaryEvolver

Built-in profiles

Profile Types Notable policies
core fact, note, goal, task, event Shared primitives all other profiles can opt into
personal + preference, observation preference → REPLACE (60d decay)
child_development + observation (tagged), milestone, concern observation tags: language/motor/emotional/cognitive/social
research_paper + claim, method, evidence, limitation, open_question evidence → REINFORCE (multiple papers corroborate)
engineering_design + decision, constraint, risk, assumption, todo decision → SUPERSEDE, risk → FLAG
legal + obligation, exception, deadline, definition, citation definition → SUPERSEDE
medical_literature + finding, population, intervention, outcome, limitation outcome → REINFORCE across studies

Custom profiles via Python dataclass, JSON, or YAML.

Storage

Three backends, one ABC:

Store Persistence Notes
InMemoryStore None Default; fastest
JSONLMemoryStore Append-only file Last-write-wins; tombstones; compact() rewrites
SQLiteMemoryStore SQLite file Indexed on (workspace, type, subject); persists embeddings; auto-migrates v0.2 → v0.4 schemas
from typedmem import SQLiteMemoryStore, DomainProfile

store = SQLiteMemoryStore.for_profile(
    DomainProfile.builtin("research_paper"),
    path="papers.db",
)

Retrieval

from typedmem import HashingEmbeddingProvider, Retriever

retriever = Retriever(store, embedder=HashingEmbeddingProvider())
hits = retriever.relevant(
    "blood pressure reduction",
    types=["evidence"],
    workspace="cardiology",
)

relevant() blends three signals: semantic (cosine), recency (exponential decay), confidence (with type-specific half-life). Without an embedder, falls back to token overlap.

Evolution

Evolvers read stored memories and produce auditable actions.

from typedmem import (
    ContradictionSurfacer, PreferenceDriftDetector,
    GoalResolver, SummaryEvolver,
    HashingEmbeddingProvider, AnthropicClient,
)

# 1. Pure read: walk the FLAG graph.
for cluster in store.contradictions():
    print(f"{len(cluster)} memories cross-link as contradictions")

# 2. Annotation: catch unstable preferences.
PreferenceDriftDetector(min_replaces=3, window_days=30).evolve(store)

# 3. Safe match: dry-run first, then commit.
embedder = HashingEmbeddingProvider()
plan = GoalResolver(embedder, threshold=0.85).evolve(store, dry_run=True)
print(plan.summary())
GoalResolver(embedder, threshold=0.85).evolve(store)            # commit

# 4. Non-destructive summary of stale events.
SummaryEvolver(AnthropicClient(), min_cluster_size=3).evolve(store)
# Originals untouched; new memory links via metadata["summarizes"].

Every action emits an EvolutionRecord (evolver, action, input_ids, output_ids, reason, timestamp) and gets appended to each affected memory's metadata["evolution_history"]. No black-box mutations.

CLI

typedmem profiles                                            # list built-in domain profiles
typedmem --profile research_paper add "..." --document-id paper.pdf
typedmem --profile engineering_design list --type decision
typedmem search "blood pressure" --type evidence
typedmem evolve --evolver contradictions
typedmem evolve --evolver goals --apply --threshold 0.9      # dry-run by default
typedmem history MEMORY_ID                                   # audit trail for one memory
typedmem workspaces

Default store: ~/.typedmem/memories.db (override with --store path.db or --store path.jsonl).

Status & roadmap

v0.4 is the first public release.

  • v0.5 sentence-transformer embedder, profile composition (extends), destructive compaction (MemoryStore.compact_summaries())
  • v0.6 hybrid BM25+semantic retrieval, query DSL, observability hooks

What TypedMemory doesn't do and doesn't plan to:

  • ship document chunkers / loaders — define the ingest() seam, bring your own (unstructured, langchain, plain regex)
  • ship its own vector DB — the abstraction is ready for one, but brute-force cosine wins under ~50k memories
  • pull network dependencies into the default install — every provider is an opt-in extra

License

MIT — see LICENSE.

Contributing

Issues and PRs welcome. Please run pytest and the demos in examples/ before opening a PR; CI runs them on Python 3.10/3.11/3.12.