becomer-agents

Multi-tenant agent memory framework powered by BECOMER.

Zero tokens per recall. Any LLM. Shared memory across agents without message passing.

pip install becomer-agents

---

Why this exists

LangChain, LlamaIndex, CrewAI, AutoGen — they all have memory. But:

• Memory dies when the process ends
• Memory is tied to one LLM provider
• Agents can't share memory without message passing
• Every recall burns 500–7,000 tokens on an LLM reasoning pass

becomer-agents fixes all four. Memory persists across sessions, works across every LLM simultaneously, agents share a namespace without coordination code, and retrieval costs zero tokens.

---

How namespaces work

Every agent gets a namespace: {task_id}.{role}

task-abc.researcher   ← researcher agent's private memory
task-abc.executor     ← executor agent's private memory
task-abc.shared       ← readable and writable by all agents

Shared memory without message passing. No state files. No coordination logic.

---

Quick start — multi-agent pipeline

import os
from becomer_agents import MultiAgentPipeline

def researcher(task, own_ns, shared_ns):
    own_ns.store("API endpoint: POST /v1/payments, OAuth2 bearer")
    own_ns.store("Rate limit: 100 req/s")
    return "Research complete"

def executor(task, own_ns, shared_ns):
    # Recall what researcher found — no message passing
    findings = shared_ns.recall("payment API endpoint and auth", top_k=5)
    own_ns.store(f"Implementation plan based on {len(findings)} findings")
    return "Plan written"

def reviewer(task, own_ns, shared_ns):
    everything = shared_ns.recall("what was found and planned?", top_k=8)
    own_ns.store("Review: APPROVED")
    return "Approved"

pipeline = MultiAgentPipeline(
    api_key=os.environ["BECOMER_API_KEY"],
    task_id="payments-task-001",
    roles=["researcher", "executor", "reviewer"],
)

results = pipeline.run(
    task="Build a payments integration",
    agents={
        "researcher": researcher,
        "executor":   executor,
        "reviewer":   reviewer,
    },
)

The terminal shows a live memory activity feed as agents store and recall.

---

Self-improving agents

from becomer_agents import SelfImprovingPipeline

pipeline = SelfImprovingPipeline(
    api_key=os.environ["BECOMER_API_KEY"],
    task_id="optimizer-001",
)

def my_agent(task, context):
    # context = what worked in previous iterations
    # use it to choose a better approach this time
    approach = "zero-shot" if not context else "few-shot + CoT"
    score = run_eval(task, approach)
    return {"approach": approach, "score": score, "output": "..."}

for i in range(5):
    result = pipeline.run_iteration(
        task="classify customer sentiment",
        fn=my_agent,
    )
    print(f"Iteration {i+1}: {result['score']:.0%} — {result['approach']}")

pipeline.finish()

Each iteration stores its outcome. The next iteration recalls what scored highest. The system compounds in intelligence across runs — zero extra tokens.

---

LangChain drop-in

from langchain.chains import ConversationChain
from langchain_openai import ChatOpenAI
from becomer_agents import BecomerMemory

chain = ConversationChain(
    llm=ChatOpenAI(model="gpt-4o"),
    memory=BecomerMemory(
        api_key=os.environ["BECOMER_API_KEY"],
        user_id="alice-123",   # per-user isolation
    ),
)

Memory persists across sessions automatically. One master key covers unlimited users via user_id.

---

Multi-tenant — one key, many users

from becomer_agents import AgentNamespace

# Alice's agent
alice = AgentNamespace(api_key, task_id="app", role="alice-123")
alice.store("Alice prefers TypeScript and dark mode")

# Bob's agent — completely isolated
bob = AgentNamespace(api_key, task_id="app", role="bob-456")
bob.recall("preferences")  # → [] — can't see Alice's memories

Isolation is enforced at the database level by BECOMER, not just application code.

---

Run the demo

Works with a BECOMER key only — no LLM keys needed:

pip install becomer-agents rich
export BECOMER_API_KEY=bcm_your-key-here
python examples/demo.py

Get a free key at becomer.net/signup.html (1,000 calls/month free).

---

Installation

# Core (no LangChain)
pip install becomer-agents

# With LangChain support
pip install "becomer-agents[langchain]"

Requires Python 3.10+.

---

Best practices

Following BECOMER's best practices gives the best recall quality:

1. Store atomic facts — one fact per store() call
2. Use specific queries — "what is the user's job title?" not "what do they do?"
3. Anchor entity names — "Alice prefers TypeScript" not "she prefers TypeScript"
4. Store signal not noise — would this be useful in 3 months?
5. Call sync() at session end — consolidates working memory
6. Use consistent user_id — different string = different namespace

---

Architecture

becomer-agents
├── AgentNamespace        task_id + role → BECOMER user_id
├── PipelineNamespaces    manages all namespaces for a run
├── MultiAgentPipeline    sequential agents with shared memory
├── SelfImprovingPipeline iterates, recalls best past approaches
├── PipelineVisualizer    rich terminal display
├── BecomerMemory         LangChain BaseMemory drop-in
└── BecomerAgentMemory    episodic + semantic memory for agents

Memory layer: BECOMER API — 94.4% LongMemEval, zero tokens per recall.

---

Links

• BECOMER API: becomer.net
• Docs: becomer.net/docs.html
• Free API key: becomer.net/signup.html
• GitHub: github.com/Becomer-net/Becomer.net

MIT License