AI agents that answer from your knowledge, on your models, with every run on record
Botlit has no non-AI half. Every answer runs your limits, draws on your own documents, and lands in a record showing which model answered and what it cost.
Most tools add a chat box to something else. Botlit's whole job is the agent: the model it runs on, the knowledge it reads, the tools it can use, the rules it obeys.
So the answer is built as a sequence you can inspect — checked before it runs, grounded in your documents, checked again before it is sent, and priced per run.
Everything below is labelled with where it actually stands in Botlit today — shipped, in progress, or on the roadmap.
AI-native by architecture, not bolted on
Four architectural choices that make AI a first-class part of Botlit rather than a chat window on the side.
Agents are the whole product
There is no non-AI half here. Bot apps, agents, knowledge, tools, rules and run history are the product — not a feature bolted onto one.
Your models, your keys, your knowledge
Agents run on providers your organisation already approved, with keys you add and keep. We resell no inference, and a model server you host works too.
Grounded before it answers
Your documents are searched while the answer is written, and the agent is told to admit it does not know rather than invent something plausible.
Every answer is checked and recorded
Your limits apply before a model is called and again to what it says. Every run — good or failed — leaves the time, cost, model and tools behind.
From intent to a governed action
Walk the path a request takes. Select any stage to see what happens there and what backs it.
Stage 1 of 5 · Before
Your rules run first
How often, how much, and what may never be discussed — all checked before a token is paid for.
The question is measured against the limits you set for that agent: how often it may run, how much it may spend today, and the words, patterns or personal data it must never handle. Over the line, the person gets a plain refusal and you are charged nothing.
What the AI in Botlit actually does
Filter by delivery status, then open any capability for the detail and what backs it. No capability is listed as shipped without something in the product behind it.
Where the AI shows up
The places AI meets the work in Botlit — and how far each one has actually got.
Chat with your agents
ShippedOne conversation, everything behind it: your limits, your documents, your tools, the model you chose, and a record once it is done.
- Chat with any agent and see which model answered and what it used
- A blocked question or answer returns one clear refusal, never half of it
- Every turn is recorded, including the ones that fail or time out
See it as a real scenario
Every AI capability above shows up in a concrete Botlit story. Open one to read the full walk-through.
Answers from your own documents — or an honest “I don't know”Your knowledge is searched as the answer is written, and the agent is told to admit a gap rather than invent something.Read the story
Agents with hands: your tools, used mid-conversationThe agent can look up an order or open a ticket while answering, then carry on — with a hard cap on how far it goes.Read the story
The router that decides which specialist takes the messageOne bot in front of many agents, dispatched by your rules or by what the message says, with a fallback when it is unclear.Read the story
Agents that hand work to other agents — with a leashA generalist passes the question to a specialist that really runs, and the answer comes back in the same conversation.Read the story
Guardrails that stop the model, not guidelines that describe itRate, spend and content limits checked before the model runs and again on its reply, with one clear refusal instead of a leak.Read the story
Every answer leaves a record you can audit and bill againstTime, cost, model, tools used and how much of your knowledge was read — for every run, including the ones that fail.Read the story
A copilot that drafts your agents and a supervisor that watches themPlanned: describe a bot and approve the draft piece by piece, then get scheduled findings about the agents you run.Read the storyAI you can actually let near your data
An AI-native product has to be governable. Here is where Botlit stands on each control — including the parts still being built.
Permissions checked on every request
Every action — including asking an agent a question — is checked against your workspace roles before it runs. The AI path gets no shortcut of its own.
Your workspace is the wall
Agents, documents, keys, conversations and run records belong to one workspace, taken from who is asking rather than what they claim. Search never leaves it.
Your limits are the ones that apply
Rate, spend and content limits come from your workspace, not our defaults. One you set is binding; one you never set changes nothing, so nothing blocks by surprise.
A trail per run, and per release
Each run keeps its question, answer, status, time, tokens, cost and model. Publishes and rollbacks are added, never overwritten, so “which version said that” is a lookup.
Hardening against poisoned input
Text from your documents, your tools and other agents still enters the prompt unfiltered, and a destructive tool call is not yet held for approval. Both are planned.
Spend alerts and a credit ceiling
Each run already records what it cost. Still missing: a view that slices spend by bot, agent and model, an alert when one drifts, and a ceiling on Botlit's own AI use.
Models
Bring your own keyNo house model, no resold inference: your workspace registers the providers you approve, keys stay encrypted, and every agent binds to one. Two honest limits — tool use and agent handoffs work on OpenAI-compatible providers only, and replies do not stream yet.
- OpenAI
- Anthropic
- Cohere
- Mistral
- Groq
- Together
- HuggingFace
- Azure OpenAI
- AWS Bedrock
- Self-hosted, via an OpenAI-compatible endpoint
The honest answers
Shipped: your limits enforced, grounding in your documents, routing between specialists, versioning with rollback, and a record of every run. Partly there: tool use and agent handoffs, on OpenAI-compatible providers only. Planned: sources under the answer, streaming replies, the copilot and the scheduled supervisor, an embeddable web chat widget for your own site, and automated human handoff when the app detects a low-confidence or escalation turn.
Put Botlit’s AI to work
Botlit has no non-AI half. Every answer runs your limits, draws on your own documents, and lands in a record showing which model answered and what it cost.