On by default, every call
getOperatingContext assembles your company profile, playbooks and knowledge, then prepends it to every callLLM and callLLMStream. Grounding is the standard behavior — not a mode you remember to switch on.
grounded AI · default-on
Grounded AI with your company knowledge
Grounded AI answers from your business, not the open web. dolv prepends your company profile, playbooks and knowledge base to every AI call by default — so output is on-brand and on-context, then turns that context into real work. Grounded in your company, not generic.
On-brand because it's grounded — not because someone rewrote it. dolv it.
the idea
A generic model guesses. Grounded AI knows you.
An off-the-shelf LLM knows the internet's average answer — plausible copy you have to rewrite to sound like your company. Grounded AI knows your business: your product, your ICP, your voice, because your company context is anchored to every response. The difference between knowledge-grounded AI and a chatbot is whether the answer fits your company before you edit it.
Three sources, one block
Your saved company profile, your playbooks, and a crawled knowledge base assemble into a single grounding block — so chat, agents and campaigns all reason from the same company context.
On-brand by construction
compose_content and the prepare_* tools draft from your grounding block, so emails, LinkedIn posts and WordPress drafts already sound like your company — not generic AI you have to rewrite.
Grounded and agentic
Grounded chat does not stop at a smarter answer — it calls 25+ tools to read live data and prepare real work, all reasoned from your business context, in the loop.
Same grounding for agents
Configurable agents and multi-agent campaigns inherit the same grounding block. Every agent run is grounded in your company context and logged to run history, capped to a $250/mo budget.
Explicit opt-out
Default-on, but never trapped: passing {grounded:false} returns a raw, ungrounded response for that single call — a clean escape hatch without changing the default anywhere else.
- 3 sources in every grounding block: profile · playbooks · knowledge
- 25 tools grounded chat can call to execute work
- 100 of LLM calls grounded by default — chat, agents, campaigns
- 20 read + write integrations feeding live, grounded reads
default-on grounding
How dolv grounds every AI call by default
Under the hood, getOperatingContext assembles your company profile,
playbooks and knowledge into one grounding block and prepends it to every
callLLM and callLLMStream. Grounding isn't a feature you
remember to enable — it's the default, which is why dolv reads as a
context-aware AI across the whole product, not a single
clever prompt.
01
Assemble
getOperatingContext pulls your company profile, playbooks and knowledge base into one consistent grounding block — the same block, every time.
02
Prepend
That block is prepended to the call before the model ever sees your request, so it reasons from your business first, not the open web average.
03
Reason
The model answers in your context — your product, ICP, voice and rules — so the output already fits your company instead of needing a rewrite.
04
Act
Grounded chat then calls real tools to draft, prepare and queue work in the Approvals inbox — grounded reasoning that turns into grounded action.
company knowledge
Your knowledge base, three sources deep
The grounding block is built from three real sources: your saved company profile, your playbooks, and a crawled knowledge base. Set them once and every grounded answer — across chat, agents and campaigns — draws from your AI knowledge base for go-to-market.
Your website becomes knowledge automatically: the built-in crawl4ai-powered engine ingests your pages (with a BFS fallback), and the Settings "Website" field auto-saves each crawl into knowledge. Go deeper on the AI knowledge base and company playbooks for AI.
on-brand content
On-brand AI content, not generic slop
Because the content tools draft from your grounding block, the output isn't generic. Emails, LinkedIn posts and WordPress drafts inherit your voice, product framing and positioning — on-brand AI content by construction.
That's the practical payoff of grounding: less time rewriting AI drafts to sound like you, because they already do. Read more on on-brand AI content, then see how the same grounding feeds funnel intelligence so the reads you act on stay tied to your real context.
the difference
Grounded AI vs RAG vs prompt-stuffing
"Grounded AI" gets used loosely. Here's the plain-language difference between dolv's always-on grounding, bolt-on RAG for business, and the old habit of pasting context into a prompt by hand.
dolv grounded AI Bolt-on RAG Prompt-stuffing
When context is added On every call, by default — opt-out is explicit ({grounded:false}). Only when a retrieval step is wired in and runs successfully. Only when a person remembers to paste it into the prompt.
What gets included Company profile + playbooks + knowledge as one assembled block. Top-matching snippets from a vector store for that one query. Whatever fits in the box that day — easy to forget half of it.
Consistency across the team Same grounding for chat, agents and campaigns — nobody opts in. Varies by how each query retrieves; tuning lives in the pipeline. Depends on the person — output drifts off-brand fast.
Does it also act? Yes — grounded chat calls 25+ tools to do real work, in the loop. No — retrieval improves an answer; it does not execute work. No — it is a one-off message, not an operating system.
Want the deep dive? Read RAG vs grounded AI or the grounded AI definition in our glossary.
who it's for
Built for teams tired of editing generic AI
If you've ever pasted your positioning into a prompt for the third time that day, or rewritten an AI draft so it stops sounding like everyone else, grounding is the fix. Point dolv at your site and connect your stack once, and every grounded answer — across AI agents, marketing automation and sales & CRM — already speaks in your company's voice.
The grounding is the same for one person or a whole team — nobody opts in, nobody forgets the context, and output doesn't drift off-brand.
questions
Grounded AI, answered
What does "grounded AI" mean?
Grounded AI means the model's answers are anchored to your company's own knowledge — your company profile, your playbooks and your knowledge base — instead of the open web's generic average. In dolv, grounding isn't a mode you switch on for special tasks: getOperatingContext assembles that company context and prepends it to every AI call, so chat, agents and campaigns all reason from your business by default.
How does dolv keep AI on-brand and on-context?
Every callLLM and callLLMStream in dolv prepends the same grounding block — company profile, playbooks and knowledge — before the model ever sees your request. Because the content tools (compose_content and the prepare_* outreach tools) draft from that block, emails, LinkedIn posts and WordPress drafts come out sounding like your company, not like generic AI. On-brand is the default state, not a checklist step.
Is grounding on by default, or do I have to turn it on?
Grounding is on by default. You don't enable it per task or remember to attach context — it is the standard behaviour for every LLM call across chat, configurable agents and multi-agent campaigns. Opting out is the explicit choice ({grounded:false}) for the rare case where you want a raw, ungrounded response. That default-on design is the core difference from bolt-on retrieval you have to wire up.
What sources feed dolv's company knowledge base?
Three sources form the grounding block: your saved company profile, your playbooks, and a knowledge base. The knowledge base is built by crawling your website — the crawl4ai-powered engine ingests your pages (with a BFS fallback), and the Settings "Website" field auto-saves those crawls into knowledge. Point dolv at your site and your docs once, and every grounded answer draws from them after that.
Is grounded AI the same as RAG?
They are related but not identical. RAG (retrieval-augmented generation) bolts a retrieval step onto a query — it fetches matching snippets when the pipeline runs. dolv's grounding is default-on: it always assembles your company profile, playbooks and knowledge for every call, the same way for chat, agents and campaigns, and that grounded context also feeds tools that execute real work. See the full breakdown on our RAG vs grounded AI page.
Can I stop the AI from using my company context for a specific task?
Yes. Grounding is default-on, but it is also explicitly optional — passing {grounded:false} produces an ungrounded response for that single call without changing the default everywhere else. So you keep on-brand, on-context output by default and still have a clean escape hatch when you genuinely want a raw answer.
See grounded AI on your own data
Point dolv at your website and connect your stack — then watch AI grounded in your company knowledge run real work, on-brand and on-context, with you in the loop. Don't just plan it. dolv it.