New: behavioral baselines on every agent

Your agent stops behaving
like itself. You know first.

ZizkaDB watches every session, builds a baseline, and flags the ones that drift. Before your users do.

Start free →Read docs
$pip install zizkadb-sdk

Works with anything

Python SDKpip
TypeScript SDKnpm
MCP Serveruvx
REST APIany language
Claude Desktopvia MCP
Cursorvia MCP
LangChainPython / JS
CrewAIPython
AutoGenPython
OpenAI AgentsPython / JS
LlamaIndexPython
Customany stack

Three moments other tools miss

LangSmith and Langfuse show you traces after you go looking. ZizkaDB watches while you sleep.

WHAT NOBODY ELSE CATCHES
01
v2 quietly regressed
Subtle drops in quality. No errors. Eventually a user complains.
ZizkaDB compares each session against the baseline. See exactly where v2 diverged.
02
Something broke this session
Wrong answer. No idea what the agent saw or why.
db.why(event_id). Full causal chain back to the user message. One line.
03
Complaint about Tuesday
Customer says you told them something wrong three days ago.
db.at(timestamp). Exact agent state at that moment. Checksummed.

Pick a stack

Python · TypeScript · MCP · HTTP. All four are the same product.

quickstart.py
from zizkadb import ZizkaDB

db = ZizkaDB("agdb_live_xxxx")   # your API key

result = await db.log(
    agent="my-bot",
    event="tool_call",
    data={"tool": "search", "query": "pricing"},
)

# Why did this happen?
chain = await db.why(result.event_id)
chain.print()
Install
pip install zizkadb-sdk
Requires
Python 3.10+
Only dependency
httpx

Three lines. That's it.

Log. Link with parent_id. Ask why.

quickstart.py
from zizkadb import ZizkaDB

db = ZizkaDB("agdb_live_xxxx")   # your API key

# 1. Log what your agent does
msg = await db.log(agent="support-bot", event="user_message",
    data={"text": "why is my bill $200?"})

# 2. Link every action to its cause
tool = await db.log(agent="support-bot", event="tool_call",
    data={"tool": "get_billing", "user": 123},
    parent_id=msg.event_id)          # ← causal link

# 3. Ask why, at any time, for any event
chain = await db.why(tool.event_id)
chain.print()
# user_message: "why is my bill $200?"     [14:32:01]
#   └── tool_call: get_billing(user=123)    [14:32:02]
#       └── response: "Found anomaly"       [14:32:03]

Already using Claude or OpenAI?

Their memory is for the chat. ZizkaDB is for your code: every decision, every cause, every trigger.

No wrappers. No monkey-patching. Just parent_id.

claude_agent.py
import anthropic
from zizkadb import ZizkaDB

db     = ZizkaDB("agdb_live_xxxx")
client = anthropic.Anthropic()

async def run(user_input: str):
    # Log the user turn
    turn = await db.log(
        agent="claude-3-5-sonnet",
        event="user_message",
        data={"text": user_input},
    )

    response = client.messages.create(
        model="claude-3-5-sonnet-20241022",
        max_tokens=1024,
        messages=[{"role": "user", "content": user_input}],
    )

    # Log Claude's reply, causally linked to the input
    await db.log(
        agent="claude-3-5-sonnet",
        event="assistant_response",
        data={"text": response.content[0].text},
        parent_id=turn.event_id,     # ← one line. full lineage.
    )
    return response
Session 1
Causal chain on every decision.
Session 10
Semantic search. Time travel.
Session 50
Behavioral baseline live. Alerts coming.

What vector DBs can't do

Vector DBs store embeddings. ZizkaDB stores what happened, why, and what came next.

🔍
Causal Lineage
Every event links to its cause. Trace any output to the input that triggered it.
Time Travel
Replay your agent's exact state at any past moment.
🧠
Semantic Search
Search history in plain English. No schema. No keywords.
🚨
Behavioral Baseline
See when an agent stops looking like itself. Investigate before users notice.
🤝
Agent Handoff
Pass only the causally relevant context. Stops full-history-in-prompt.
📋
Audit Log
SHA-256 checksummed. Append-only. Export a signed trail anytime.

How it compares

Facts only. Verify every row in their docs.

LangSmith · Mem0 · Pinecone · ZizkaDB · May 2026

CapabilityLangSmithMem0PineconeZizkaDB← us
Agent event logging
Causal lineage (parent → child)~
Time travel: state at timestamp
Semantic search over history
Works with any framework/model~
Behavioral baseline per agent
Cross-agent fleet queries
Tamper-evident audit export
Self-hostable for free

~ = partial support  ·  LangSmith works mainly within the LangChain ecosystem

Pricing

Free if you self-host. Paid if you don't want to.

Self-Hosted
Freeforever
  • Full feature set
  • Your infra
  • Docker Compose
  • Community support
Setup guide →
MOST POPULAR
Pro
€29per month
  • 100M events
  • 90-day retention
  • 3 projects
  • Email support
Start 14-day free trial →
No credit card needed
Team
€99per month
  • 1B events
  • 1-year retention
  • 10 seats
  • Priority support
Start 14-day free trial →
No credit card needed

Run it yourself

Full feature set. No account. No limits. Forever.

$ git clone https://github.com/Zizka-ai/agentdb
$ cp .env.example .env && nano .env
$ docker-compose -f infra/docker-compose.yml up
Setup guide →