GitHub - bagofwords1/bagofwords: Chat with your data - with memory, rules, and observability built in. Deploy in 2 minutes

Features

• Analysis: Create reports and dashboards, generate queries, and run deep or root cause analysis.
• Agent context: Configure each agent with the right data, tools, credentials, instructions, permissions, and starters.
• Query acceleration: Speed up analytical queries 10–1000× with an auto-updating cache that spares the source.
• Automations: Schedule reports, run recurring tasks, and trigger investigations from events and webhooks.
• Channels: Run headlessly via Claude Code, Codex, and other MCP clients, or through Microsoft Teams, Slack, WhatsApp, email, Excel, and the web app.
• MCP gateway: Connect agents to MCP servers and custom APIs, then expose their context and tools to MCP clients through one governed gateway.
• Evals and self-improvement: Set evals for expected behavior. When they fail, agents can draft instruction fixes and re-run the evals; passing changes can wait for approval or be promoted automatically.
• Governance: Control access with RBAC, approvals, audit logs, service accounts, SSO, model policies, and row-level security support.

Deploy anywhere

Quick Start

# Run with SQLite (default)
docker run -p 3000:3000 bagofwords/bagofwords

Run with PostgreSQL

docker run -p 3000:3000 \
  -e BOW_DATABASE_URL=postgresql://user:password@localhost:5432/dbname \
  bagofwords/bagofwords

Docker Compose and Kubernetes deployments are also available and recommended for servers. See the installation docs.

From Analysis Context to Action

Start with chat

Ask questions across your data and get queries, charts, reports, dashboards, deep analysis, and root cause analysis. Agents plan their work, use tools, and reflect on the result.

Give every agent the right context

Choose the data and tools an agent can use, provide the right credentials, and add instructions and starters for its job. Manage shared business definitions and guardrails with versioning, review flows, and Git sync for dbt, markdown, code, and more.

Turn analysis into repeatable work

Schedule reports and investigations, trigger agents from webhooks, and send results to the people and systems that need them. The same analysis context is available in chat, automations, channels, and external MCP clients.

Evaluate and improve

Define eval sets for the behavior that matters and run them as agents change. When an eval fails, a self-improving loop can draft instruction changes and re-run the tests up to a configured limit. Passing candidates can wait for approval or be promoted automatically.

Monitor every run

Get full observability across every message, query, plan, tool call, result, latency, cost, and failure. LLM-as-judge scores for accuracy, instruction coverage, and context use help teams catch unreliable data results, diagnose what happened, and feed failures back into evals and self-improving loops.

Agents pane: build and manage data agents at scale

Create agents for teams, systems, and business domains. Scope their data, tools, credentials, instructions, permissions, starters, evals, and channels; review pending changes and manage the full agent fleet from one pane.

Speed up analytical queries with an auto-updating cache

Query acceleration with caching lets agents query a warehouse or database at speed, improving the experience for users while reducing load and cost on the source. Admins choose what to cache — whole tables or specific queries — and BOW keeps it refreshed on a schedule.

A refresh is a full run of your SQL, so a copy refreshed more often than it is queried costs more than it saves — the break-even is about 1.8 agent questions per refresh on BigQuery and 1.0 on Snowflake. One copy is shared by every agent that activates it, and row-level security can filter it per person against their profile attributes, groups, or roles. See the caching documentation.

Architecture

Bag of Words sits between your models, enterprise data, tools, and channels. It builds a governed analysis context for each agent, then carries that context from an interactive question to a scheduled or event-driven workflow.

Bring your own models and infrastructure. Connect databases, warehouses, BI systems, files, business apps, MCP servers, and custom APIs without tying the workflow to one provider.

Integrations

Bring Any LLM

Use your own API keys, endpoints, and model deployments. Multiple providers and models can be configured in the same environment.

Connect Any Data

Connect Tools Through MCP

Bag of Words can connect to any MCP server or custom API. Ready-to-connect MCP integrations include:

Browse the Web

Add a Browser connection to let an agent open web pages in a real (headless) browser — navigate, read page content, interact with elements, and capture screenshots — scoped to a URL allowlist you define. Useful for data that only lives behind a UI, including internal portals reachable from your deployment. The browser tools appear for an agent only when a browser connection is attached to it.

Run Anywhere

Enterprise

For teams that need stronger security, compliance, and governance:

• Self-hosted: Deploy on your own infrastructure and keep control of your data.
• SSO and provisioning: Connect Google Workspace and OIDC-compatible identity providers, with SCIM and LDAP support.
• RBAC: Apply fine-grained permissions to agents, data, tools, and administration.
• Approvals and audit: Review changes and track agent and data operations.
• Service access: Use API keys and service accounts for headless workflows.
• Model controls: Decide which providers and models are available to each organization.

Security and Privacy

Bag of Words captures basic usage statistics from self-hosted instances to help improve the product. Disable telemetry in bow-config.yaml:

telemetry:
  enabled: false

You can also disable Intercom support chat: