Governed context for AI agents

Connect your AI agentsto governed context.

One versioned, human-approved corpus of documents and business definitions. Served over MCP and REST with typed citations. When the corpus cannot support an answer, the system abstains.

Request access Read the docs

Same corpus and rules for humans and agents. API + SDK available today.

Trusted by teams building production agents

# Connect any MCP-compatible client
claude mcp add --transport http renbase https://api.renbase.ai/mcp \
  --header "Authorization: Bearer kb_live_..."
exact definition resolution hybrid search citations abstention

Full REST API and SDK available for custom integrations.

01 · The problem

Agents fail on missing business context, not on the model

Production data agents receive questions they cannot answer correctly because the required context is not available in a governed form.

  1. 01

    Undefined business terms

    Revenue can mean ARR, run-rate, or recognized revenue. The agent has no approved definition to resolve against.

  2. 02

    Ambiguous sources of truth

    Multiple tables or views with similar names exist. The agent cannot determine which one is authoritative.

  3. 03

    Tribal and conditional knowledge

    Rules such as “USCAN deals after 2025 live in Affinity” exist only in Slack, PDFs, or people’s heads. They are never applied consistently.

  4. 04

    No governance on what the agent sees

    Managed RAG systems ingest documents automatically. Unapproved, outdated or conflicting content reaches the model without human review.

02 · The product

Renbase is the governed corpus your agents query

One corpus holds the ingested documents and the business definitions. Definitions are versioned, and nothing influences an answer until a human approves it. Agents and humans query that corpus under the same rules and the same usage accounting.

01

Ingested documents

PDF, DOCX, PPTX, XLSX, HTML, Markdown and plain text land in one hybrid index, chunked and retrievable next to the definitions.

02

First-class business definitions

Metrics, entities, rules and glossary terms are objects, not passages. Asking for one by name returns that exact entry, not something similar.

03

Versioned and approved

Every edit creates a new immutable version. Approver and last verification travel with the entry, and stale sources are flagged.

Access today Workspace (web UI) REST API SDK MCP (for agents)

03 · Answer contract

Three non-negotiable rules

The same three rules apply in the workspace and over MCP. They are properties of the system, not instructions in a prompt.

Rule 01 — Sources always

No answer without a source

Every claim is anchored to an approved definition or an ingested document. Version, approver and last verification travel with the citation.

Rule 02 — Human approval

Drafts never serve answers

Machine-generated or imported definitions land as drafts. They remain invisible to the answer engine until a team member approves them.

Rule 03 — Silence over invention

Abstain when unsupported

Missing, expired or contradictory content produces an explicit abstention, or surfaces both conflicting sources with provenance. The system does not invent or silently choose.

04 · How it works

Capture → Govern → Serve

  1. 01

    Capture

    Upload documents via workspace, REST API or SDK. Supported formats today: PDF, DOCX, PPTX, XLSX, HTML, Markdown and plain text. dbt artifacts import as draft definitions, and a client-side CLI can propose entities from your warehouse schema without sending credentials or row data.

  2. 02

    Govern

    All machine-produced content starts as draft. Team members approve, edit or reject. Edits create immutable new versions, feedback on a bad answer can be promoted into a rule, and stale sources are flagged — freshness is attached to every response.

  3. 03

    Serve

    Agents and the workspace query the same corpus over MCP or REST. Exact name resolution for approved definitions, hybrid retrieval for documents and free text. Every response carries typed citations or an abstention reason.

Coming soon

  • Native connectors: Google Drive, SharePoint, Salesforce, Confluence, Notion and others
  • Direct database connections (warehouse and operational DBs)

Connectors and database links will produce draft definitions and documents for human approval before they enter the corpus.

05 · Integrations

API, SDK and upcoming connectors

Available today
  • Full REST API
  • Official SDK
  • MCP endpoint — Claude, Cursor and any MCP client
  • Workspace UI
  • dbt artifact import (manifest + catalog)
  • Warehouse schema introspection (client-side CLI)
Coming soon
  • Google Drive · SharePoint · Confluence · Notion
  • Salesforce and other business systems
  • Databases and data warehouses
  • Scheduled re-sync against the source

All incoming content, from connectors or databases, lands as drafts and requires human approval before influencing answers.

06 · Agent interface

MCP tools

Any MCP-compatible client connects with a per-organization API key. The tools enforce the same answer contract as the human workspace.

get_definition(name)
Exact resolution of an approved definition by name or alias. Returns body, version, approver and last verification. Surfaces conflicts when present.
Cost Free
list_entities(kind?)
Lists approved definitions in the organization corpus.
Cost Free
search_context(query)
Hybrid retrieval across documents and definitions.
Cost 1 credit
expand_source(chunk_id)
Expands a citation into full surrounding context.
Cost Free
ask(question)
Full answer with citations and abstention support.
Cost 1 credit 
# Connect any MCP-compatible client
claude mcp add --transport http renbase https://api.renbase.ai/mcp \
  --header "Authorization: Bearer kb_live_..."

One command in any MCP client. The key scopes the corpus to your organization.

07 · Differentiation

Renbase vs managed RAG

Managed RAG (Bedrock Knowledge Bases, Vertex AI RAG, Azure AI Search) indexes documents and returns relevant passages. Renbase treats the answer as a contract: cited or abstained, with human-approved definitions as first-class objects.

Core promise
Managed RAG Relevant passages + generated answer
Renbase Cited answer or explicit abstention
Content model
Managed RAG Documents in a vector index
Renbase Documents + governed definitions in one corpus
“What does X mean here?”
Managed RAG Similarity search
Renbase Exact deterministic resolution with version and approver
Human approval gate
Managed RAG None — connectors sync automatically
Renbase Machine content is draft until approved
Conflicting sources
Managed RAG Model picks or blends
Renbase Both returned, marked conflicting, with provenance
Outdated sources
Managed RAG Retrieved like any other
Renbase Flagged stale; answers include freshness
Agent access
Managed RAG Retrieval APIs
Renbase MCP tools under the same contract as humans
Deployment
Managed RAG Bound to one cloud or ecosystem
Renbase Neutral SaaS or Enterprise BYOC inside your infrastructure

08 · Proof

Built for production agents

  • Multi-tenant by design

    Isolated corpus, keys and usage accounting per organization, from the first commit.

  • One contract for humans and agents

    The workspace and the MCP tools answer under the same rules. There is no relaxed agent path.

  • Typed citations on every claim

    Source, version, approver and last verification, attached to the claim rather than appended as a list.

  • Explicit abstention

    Missing, expired or conflicting context is reported as such. Neither definition is chosen silently.

  • REST API and SDK available today

    Ingestion, retrieval, governance and usage are all reachable programmatically.

  • Enterprise BYOC available

    The full stack runs inside your infrastructure. Corpus, questions and answers never leave it.

09 · Deployment

Two options

Multi-tenant SaaS

Option A

Each organization has an isolated corpus, members, API keys and usage. Members sign in with email and a one-time code — no passwords.

  • Credits consumed per question. No seats, no subscription
  • Workspace, REST API, SDK and MCP included
  • Per-organization isolation, keys and rate limits from day one
Request access

Enterprise BYOC

Option B

The full stack runs inside your cloud or on-premises. Corpus, questions, answers and usage never leave your infrastructure.

  • Deployment of the Renbase stack guided by our team
  • Knowledge-management onboarding for the corpus and glossary
  • Custom limits and support for your organization
Contact us

10 · FAQ

Frequently asked questions

01 What is Renbase?

Renbase is a governed corpus that AI agents query. It holds ingested documents and first-class business definitions — metrics, entities, rules and glossary terms — in one hybrid index. Definitions are versioned and a human approves each one before it can influence an answer. Agents connect over MCP or REST and get typed citations, or an explicit abstention when the corpus cannot support the answer.

02 How is this different from managed RAG?

Managed RAG indexes documents and returns relevant passages; the model then writes something. Renbase treats the answer as a contract: every claim is cited with version and approver, or the system abstains. Definitions are objects that resolve exactly by name, not passages found by similarity. And nothing a machine produced serves answers until a person approves it.

03 What does “abstention” mean in practice?

When the corpus has no approved content for the question, when the relevant entry is expired, or when two approved definitions conflict, the answer is not generated anyway. The system returns an explicit abstention with the reason, or returns both conflicting definitions with their provenance and scope. It does not pick one silently.

04 How do agents connect?

Over MCP (Streamable HTTP) or REST, with an API key per organization. get_definition resolves a name or alias to the approved entry. list_entities lists what is approved. search_context runs hybrid retrieval across documents and definitions. expand_source opens a cited passage into its full context. ask returns a complete cited answer. get_definition, list_entities and expand_source are free; search_context and ask cost one credit.

05 What can I ingest today, and what is coming?

Today you upload documents through the workspace, the REST API or the SDK: PDF, DOCX, PPTX, XLSX, HTML, Markdown and plain text. Native connectors (Google Drive, SharePoint, Salesforce, Confluence, Notion and others) and direct database connections are coming. Everything they produce will land as drafts for human approval before entering the corpus.

06 Does Renbase touch my warehouse or its credentials?

No. Renbase serves context; your agent acts. It does not run or generate SQL against your warehouse, and warehouse credentials never leave your side. If you want table names proposed as entities, schema introspection runs as a client-side CLI that submits only names and descriptions, as drafts, for your team to review. No row data is sent.

07 Where does the data run?

Two options. Multi-tenant SaaS, where each organization has an isolated corpus, keys and usage accounting, and credits are consumed per question. Or Enterprise BYOC, where the full stack runs inside your cloud or on-premises and the corpus, questions, answers and usage never leave your infrastructure.

Give agents the context they need to stop failing.

Create an organization or deploy BYOC. Same product, same answer contract. API and SDK available today. Native connectors and database connections coming soon.

Contact us Documentation