How do I install BanditDB?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does BanditDB need an API key?

The publisher lists these environment variables: BANDITDB_URL, BANDITDB_API_KEY.

What transport does BanditDB use?

BanditDB runs locally over stdio.

When was BanditDB last checked?

Last checked 2026-11-08. Persistent decision memory for agents — learns which action works in which context

BanditDB MCP server

BanditDB Python SDK

The official Python client and Model Context Protocol (MCP) server for BanditDB — the ultra-fast, lock-free Contextual Bandit database written in Rust.

BanditDB abstracts away the complex linear algebra of Reinforcement Learning (LinUCB, Thompson Sampling) behind a dead-simple API. Build real-time personalizers, dynamic A/B tests, and give LLM agents mathematically rigorous persistent memory.

mcp-name: com.banditdb/mcp

Installation

pip install banditdb-python

Requires the BanditDB Rust server running (default: http://localhost:8080).

1. Standard SDK Usage

The client features automatic connection pooling, exponential backoff retries, and strict timeouts.

from banditdb import Client, BanditDBError

# Connect to the BanditDB server.
# Pass api_key if BANDITDB_API_KEY is set on the server.
db = Client(
    url="http://localhost:8080",
    timeout=2.0,
    api_key="your-secret-key",   # omit if server runs without auth
)

try:
    # 1. Create a campaign (run once at startup)
    # algorithm defaults to "linucb"; use "thompson_sampling" for Bayesian exploration
    db.create_campaign(
        campaign_id="checkout_upsell",
        arms=["offer_discount", "offer_free_shipping"],
        features=["is_mobile", "cart_value", "is_returning_user"],
    )
    # or: db.create_campaign(..., algorithm="thompson_sampling")

    # 2. A user arrives — ask the database what to show them.
    # Scale each value to 0–1; the SDK puts them in order, adds a constant 1
    # and normalises the vector.
    arm_id, interaction_id = db.predict("checkout_upsell", {
        "is_mobile": 1.0,
        "cart_value": min(cart_total / 200, 1.0),
        "is_returning_user": 0.0,
    })
    print(f"Showing: {arm_id}")  # e.g., "offer_free_shipping"

    # 3. The user clicked — send the reward
    db.reward(interaction_id, reward=1.0)

except BanditDBError as e:
    print(f"Database error: {e}")

Named features or a vector. With features=[...] the campaign stores the feature order in its metadata, and predict, batch_predict and interact accept a dict keyed by name. A missing or misspelt name raises ValueError before any request is sent. Campaigns created in the BanditDB console accept dicts too. The order is read once per campaign with campaign_info() and cached on the client. To manage the vector yourself, pass feature_dim instead and send lists — normalize_context() scales one to unit length.

All Client methods

Health

Method	Description
`health()`	Returns `True` if the server is reachable and the WAL writer is healthy.
`health_detail()`	Returns the full health dict including per-campaign `entropy` and `status` (`"ok"` / `"warning"` / `"critical"`).

Campaigns

Method	Description
`create_campaign(campaign_id, arms, feature_dim=None, alpha=1.0, algorithm="linucb", metadata=None, decay_half_life_hours=None, pacing=None, features=None)`	Register a new campaign. Pass `features` (names, in order) to predict with dicts, or `feature_dim` to send vectors. `algorithm` accepts `"linucb"`, `"thompson_sampling"`, `NeuralLinUCBConfig`, `NeuralTSConfig`, or `ProgressiveConfig`. `metadata` is an arbitrary JSON dict (≤ 64 KB). `pacing` takes a `PacingConfig` — see Budget constraints.
`list_campaigns()`	Returns a list of all campaigns (active and archived) with `alpha`, `arm_count`, and `algorithm`.
`campaign_info(campaign_id)`	Returns full per-arm state: `theta`, `theta_norm`, prediction and reward counters. Raises `APIError` (404) if not found.
`report(campaign_id)`	Business-level convergence report. `converged=True` means one arm has a statistically significant lead at 95% CI — safe to stop. `converged=False` means leading but CIs still overlap. `converged=None` means not enough data yet (< 30 rewards per arm).
`diagnostics(campaign_id)`	Operator diagnostics: per-arm theta norms, A_inv uncertainty bounds, entropy health (`selection_entropy`, `entropy_status`, `entropy_trend`, `likely_cause`, `suggested_action`), tournament traffic, and neural buffer size.
`archive_campaign(campaign_id)`	Soft-delete: pauses predictions/rewards but preserves all learned weights. Recoverable with `restore_campaign()`.
`restore_campaign(campaign_id)`	Restore an archived campaign to active status with all weights intact.
`delete_campaign(campaign_id)`	Permanently delete a campaign. Returns `False` if not found.

Predict & Reward

Method	Description
`predict(campaign_id, context, eligible_arms=None, exclude_arms=None)`	`context` is a dict of feature values (named campaigns) or a vector. Returns `(arm_id, interaction_id)`. Pass `interaction_id` to `reward()` to close the loop. The optional filters narrow candidates for this call only.
`batch_predict(predictions)`	Predict for up to 100 campaign/context pairs in a single round-trip. Each item: `{"campaign_id": str, "context": dict
`reward(interaction_id, reward)`	Record outcome. `reward` must be in `[0.0, 1.0]`. Raises `APIError` if the interaction has already been rewarded or has expired (default TTL: 24 h).

Data & Export

Method	Description
`checkpoint()`	Flush WAL, snapshot models, write Parquet shards, run neural retrain + tournament eval, rotate WAL. Returns a summary string.
`export()`	List Parquet export shards grouped by campaign. Returns `{export_dir, shards}`.

Budget constraints

Server 2.3.0+. Give a campaign one or more budgeted resources and per-arm costs; the server learns a shadow price per resource and steers spend to last the horizon, choosing costly arms only where their uplift is worth it.

from banditdb import Client, PacingConfig, ResourceConstraint

db.create_campaign(
    "promos", ["none", "discount_10", "discount_20"], feature_dim=4,
    pacing=PacingConfig(resources=[
        ResourceConstraint(
            name="promo_budget",
            budget=5_000.0,      # units available over the horizon
            horizon=100_000,     # decisions the budget should last
            arm_costs={"discount_10": 1.0, "discount_20": 2.0},  # "none" costs 0
        ),
    ]),
)

# Arms added later declare their own costs
db.add_arm("promos", "discount_30", costs={"promo_budget": 3.0})

for r in db.pacing("promos")["resources"]:
    print(f"{r['name']}: {r['utilization']:.0%} spent, lambda={r['lambda']:.3f}")

Worth knowing:

Budget is spent at predict, not reward. A recommendation you never act on still consumes budget.
Budget never resets and pacing is immutable. Once spent, costly arms stay unselectable until the campaign is recreated.
arm_costs keys must be campaign arms. The server would silently treat a misspelt arm as free; the SDK raises ValueError instead.
Predicts on a pacing campaign can return 503 if the server's WAL is saturated.
Downgrading the server below 2.3.0 silently drops pacing.

2. The AI "Hive Mind" (Model Context Protocol)

Standard LLM agents are stateless — if they route a task to the wrong model and fail, they repeat the same mistake tomorrow. BanditDB's built-in MCP server gives the entire agent swarm shared persistent memory.

Starting the MCP server

# Set environment variables before starting
export BANDITDB_URL=http://localhost:8080
export BANDITDB_API_KEY=your-secret-key   # omit if server runs without auth

banditdb-mcp

Connecting to Claude Desktop

Add to your Claude configuration file:

Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "banditdb": {
      "command": "banditdb-mcp",
      "args": [],
      "env": {
        "BANDITDB_URL": "http://localhost:8080",
        "BANDITDB_API_KEY": "your-secret-key"
      }
    }
  }
}

The agent swarm now has twelve tools:

Tool	What it does
`create_campaign`	Create a new decision campaign. Accepts `algorithm` (`"linucb"` or `"thompson_sampling"`), `alpha`, and optional `budgets` for actions that cost something limited. Use Thompson Sampling for natural Bayesian exploration with no tuning needed.
`list_campaigns`	List all active campaigns (shows `algorithm` and `alpha`) — useful to check what exists before calling `get_intuition`.
`campaign_diagnostics`	Inspect per-arm learning state: `theta_norm`, prediction counts, reward rates, and entropy health. Use when a campaign doesn't seem to be learning or one arm is dominating.
`campaign_report`	Business-level convergence report. Tells you whether the campaign has statistically converged and which arm is winning with confidence intervals.
`get_intuition`	Ask BanditDB which arm to pick for a given context. Optional `exclude_arms` / `eligible_arms` narrow the choice for this request only. Returns the arm and an `interaction_id` to save.
`batch_get_intuition`	Get decisions for multiple campaigns in a single round-trip. Pass a list of `{campaign_id, context}` dicts.
`record_outcome`	Report whether the chosen action succeeded (1.0) or failed (0.0). Updates the shared model.
`archive_campaign`	Soft-delete a campaign. Pauses predictions/rewards but preserves all learned weights.
`restore_campaign`	Restore an archived campaign to active status with all weights intact.
`campaign_budget`	Budget spend per resource: consumed, remaining, utilization, shadow price, exhaustion.
`add_arm`	Add an action to a live campaign, optionally warm-started from existing arms and with budget costs.
`set_arm_status`	Pause, retire, or reactivate an action. It keeps everything it learned.

Every decision made by any agent in the network improves the routing for all future agents.

3. Data Science & Offline Evaluation

BanditDB event-sources every prediction and reward to a Write-Ahead Log (WAL). Calling checkpoint() compiles completed prediction→reward pairs into Snappy-compressed Parquet files — one per campaign — for offline analysis with Polars or Pandas.

Every prediction is guaranteed to appear in the Parquet file even if its reward arrives hours later: BanditDB re-emits in-flight interactions at each checkpoint so delayed rewards are always captured in a future cycle.

# Checkpoint: snapshot models, write Parquet, rotate the WAL.
# Call this on a schedule or after significant traffic.
summary = db.checkpoint()
print(summary)
# "Checkpoint written and WAL rotated: 2 campaigns, offset 4821 bytes,
#  150 interactions exported, 3 in-flight re-emitted"

# List which Parquet files are available
print(db.export())
# 'Parquet files in /data/exports: ["llm_routing.parquet"]'

# Load directly from the mounted volume into Polars.
# Flat schema: interaction_id | arm_id | reward | predicted_at | rewarded_at | propensity | feature_0 | ...
import polars as pl
df = pl.read_parquet("/data/exports/llm_routing.parquet")
print(df.head())
print(df.columns)

Offline Policy Evaluation (OPE)

The SDK ships three OPE estimators in banditdb.eval. They answer the question: "what would my average reward have been under a different policy — without running a live experiment?"

Install the eval dependencies:

pip install "banditdb-python[eval]"

Estimator	Function	How it works	When to use
Replay	`replay(df)`	Accepts each interaction with probability `(1/K) / propensity` (Li et al. 2010). Unbiased sample of the uniform random policy.	Sanity check baseline. Low coverage is expected — ~1/K of interactions are used.
IPS / SNIPS	`ips(df, clip=10.0)`	Uses every interaction with importance weight `(1/K) / propensity`. Self-normalised to reduce variance. Weight clipping (default 10×) controls the bias-variance tradeoff.	Primary estimator. Use when you have enough data but want full coverage.
Doubly Robust	`doubly_robust(df, clip=10.0)`	Fits a linear reward model, then applies an IPS correction on residuals. Consistent if either the reward model or the propensities are correct.	Best statistical efficiency. Use when comparing multiple policies or sweeping `alpha`.

All three estimators:

Accept a Polars or pandas DataFrame loaded from a BanditDB Parquet export
Evaluate the uniform random policy as the target (the unbiased baseline to beat)
Raise ValueError for Thompson Sampling campaigns (propensity column is null — TS does not log propensities)
Return an OPEResult with estimate, std_error, n_used, n_total, and method

import polars as pl
from banditdb.eval import replay, ips, doubly_robust

df = pl.read_parquet("/data/exports/llm_routing.parquet")

# How much reward would a uniform random policy have earned?
print(replay(df))
# OPEResult(method='replay', estimate=0.4821, std_error=0.0312, coverage=22.1% [33/149])

print(ips(df))
# OPEResult(method='ips', estimate=0.5103, std_error=0.0187, coverage=100.0% [149/149])

print(doubly_robust(df))
# OPEResult(method='doubly_robust', estimate=0.5219, std_error=0.0141, coverage=100.0% [149/149])

# Compare against the observed reward of the logging policy:
print("Observed (logging policy):", df["reward"].mean())
# If observed >> estimate, the campaign has learned something real — it outperforms random.

Practical use: sweep alpha offline before deploying. Train a campaign on real traffic, checkpoint to Parquet, then replay different alpha values through doubly_robust() to find the best exploration level — no live experiment needed.

Note: OPE requires the propensity column, which is only written for LinUCB campaigns. Thompson Sampling campaigns log null propensities because TS arm selection is stochastic and propensity scoring requires a deterministic logging policy.

Choosing an Algorithm

BanditDB supports five algorithms, selected at campaign creation time.

Algorithm	`algorithm` value	Exploration style	When to use
LinUCB	`"linucb"` (default)	Deterministic UCB bonus: `θ·x + α√(x·A⁻¹·x)`	Predictable, tunable. Sweep `alpha` offline to calibrate.
Linear Thompson Sampling	`"thompson_sampling"`	Samples θ̃ ~ N(θ, α²·A⁻¹), scores by θ̃·x	Bayesian posterior — no alpha-sweep needed. Concurrent users automatically diversify choices.
NeuralLinUCB	`NeuralLinUCBConfig(...)`	Deep MLP embedding + LinUCB in embedding space	Non-linear reward functions. Retrains the MLP every N rewards.
NeuralTS	`NeuralTSConfig(...)`	Same MLP embedding as NeuralLinUCB; scores by a posterior draw in embedding space instead of a UCB bound	Non-linear rewards where long-run performance matters more than early regret. Converges further at 20K+ interactions.
Progressive	`ProgressiveConfig(...)`	Self-tuning tournament: runs base + challenger in parallel, shifts traffic to the winner	Zero-configuration model selection. Picks the best algorithm automatically.

from banditdb import Client, NeuralLinUCBConfig, NeuralTSConfig, ProgressiveConfig

db = Client("http://localhost:8080")

# LinUCB (default)
db.create_campaign("routing", ["fast", "cheap"], feature_dim=4, alpha=1.5)

# Thompson Sampling — natural Bayesian exploration, alpha=1.0 is ideal
db.create_campaign("routing_ts", ["fast", "cheap"], feature_dim=4,
                   algorithm="thompson_sampling")

# NeuralLinUCB — learns a deep embedding of the context, then applies LinUCB
cfg = NeuralLinUCBConfig(
    context_dim=4,     # must match feature_dim
    embed_dim=32,      # arm matrix dimension (default 32)
    hidden_dim=128,    # MLP hidden layer width (default 128)
    retrain_every=200, # retrain the MLP every N cumulative rewards
)
db.create_campaign("routing_neural", ["fast", "cheap"], feature_dim=4, algorithm=cfg)

# NeuralTS — same parameters as NeuralLinUCB, posterior-sampling exploration
cfg = NeuralTSConfig(context_dim=4, embed_dim=32)
db.create_campaign("routing_nts", ["fast", "cheap"], feature_dim=4, algorithm=cfg)

# Progressive — runs LinUCB vs NeuralLinUCB, shifts traffic to whoever wins SNIPS checkpoints
cfg = ProgressiveConfig(
    base="linucb",
    challenger=NeuralLinUCBConfig(context_dim=4, embed_dim=32),
    min_obs=100,       # minimum buffer entries per arm before any traffic shift
    required_wins=3,   # consecutive checkpoint wins to earn one traffic step
    step_bps=1000,     # traffic delta per win run, in basis points (1000 = 10%)
)
db.create_campaign("routing_prog", ["fast", "cheap"], feature_dim=4, algorithm=cfg)

All five algorithms share the same predict → reward loop. Either neural config also works as a Progressive base or challenger.

NeuralLinUCB vs NeuralTS

	NeuralLinUCB	NeuralTS
Exploration	UCB upper bound (deterministic)	Posterior sample (stochastic), spread scaled by `alpha`
Early regret	Lower	Higher
Convergence at 20K+ interactions	Slower	Faster
Best for	Production safety, costly early mistakes	Maximum long-run performance
Paper	Xu, Wen, Zhao & Gu, ICLR 2022 (arXiv:2012.01780)	Zhang, Zhou, Li & Gu, ICLR 2021 (arXiv:2010.00827)

Both neural algorithms need a server built with the neural feature: check "neural" in db.server_info()["features"].

Error Handling

Exception	When raised
`BanditDBError`	Base exception — catch this to handle all SDK errors.
`ConnectionError`	Server is offline or unreachable.
`TimeoutError`	Request exceeded the configured timeout.
`APIError`	Server returned an error (e.g., campaign not found, unauthorized).

BanditDB

Add this server

Host a copy with our keys and logs.

Server facts

Tools

Questions