memora-crewai

Poison-resistant, Byzantine-fault-tolerant shared memory + robust voting for CrewAI crews.

CrewAI agents coordinate through tools. memora-crewai hands your crew a set of tools backed by Memora: agents remember_shared / recall_shared facts in one Byzantine-tolerant store, and crew_vote / crew_consensus on numeric estimates with a robust aggregate. Concurrent writers merge (CRDT, nothing lost), and a rogue or prompt-injected agent can't silently corrupt what the crew reads — or skew the vote.

pip install memora-crewai

One command — pulls in the memora-swarm engine and crewai. Free key (25,000 ops) at memora.optitransfer.ch/dashboard.

Quickstart

from crewai import Agent
from memora_crewai import MemoraCrewMemory

mem = MemoraCrewMemory("research-crew", api_key="opti_sk_...", node_id="planner")

researcher = Agent(
    role="Researcher",
    goal="Find and share verified facts with the crew",
    tools=mem.tools(),        # remember_shared, recall_shared, crew_vote, crew_consensus
    # ...
)

Every agent built with mem.tools() on the same room shares one poison-resistant memory.

The tools

Why it's safe under a crew

• No lost facts. Writes merge deterministically (CRDT) — concurrent agents can't clobber each other.
• No silent rewrite / no skewed vote. Every contribution is signed and attributable in the Memora engine; a rogue agent's write or an outlier vote is convicted/down-weighted, not trusted on faith.
• Durable + replayable via a local append-only log.

Pricing

First 25,000 ops free (unlimited agents), then metered per semantic op — see memora.optitransfer.ch. Reads and keepalive are never billed.

Links

• Engine & docs — https://memora.optitransfer.ch
• Source & other integrations — https://github.com/mgillr/memora-swarm
• Core package — memora-swarm

Apache-2.0 · a product of optitransfer.ch