AI needs a database. Your team needs a console. Konsil is both. Agents read the records your people see, draft the work and propose changes. People decide what goes out.
It reads your records, drafts the exact change and writes down why. Here, a valuation agent moves a target to the next stage.
Every proposal, from any agent, lands in the Review queue. Approve, Edit or Reject, one at a time.
Client emails, posts and deal stage moves are locked to a person. A few low-risk changes can auto-approve, but only if you turn that on.
Built into each area. Examples: the Signal Scanner in Marketing, the Health Agent in Customer Success, the Valuation Agent in an M&A Workspace.
Made in the Agent Builder in plain language. Read only, or allowed to propose actions for you to approve.
Connected over MCP with a key you create. It sees what that key allows, and nothing else.
Connect your own Claude with a key. Each key carries only the permissions you give it.
Look things up. Nothing changes.
Draft it. A person decides.
Each one is a separate permission on the key.
MCP is the open standard AI assistants use to connect to software. Konsil’s MCP tools call the same actions the console uses. There is no MCP tool that launches a campaign, sends an email or writes to HubSpot. Operations and the Agent Builder aren’t open to MCP yet. We open more of the console one area at a time.
Describe the job in plain language and the builder drafts the agent with you. Proposing is open today for Marketing drafts and for HubSpot tasks and property updates. Agents elsewhere start as Read only.
Talk it through with the builder. A live draft fills in as you go: the goal, what it reads, when it runs and what it may do. Prefer a form? Switch to Edit as form.
The builder suggests sources. Only the ones you approve are saved: your business background, what a team has learned, and the records it may read.
Pick Read only or Propose actions (you approve). Then set how careful it is and how much it may run.
Test with model runs it on your own data and shows what it would hand back. The banner says it plainly: Test run. Nothing was sent or changed.
Publish with a version note when you are happy. The agent’s page shows its versions, a Run now button and its run history.
Describe the job and approve its sources.
The builder drafts the goal, sources and schedule from the conversation.
Test it on your own data. Adjust what it reads.
Test runs change nothing and send nothing.
Publish version 1 with a note.
It runs on its schedule, inside its limits for runs a day and cost per run.
Approve, Edit or Reject what it proposes.
Its proposals wait in the Review queue with the reason for each.
Only what you approve.
KindsEvery agent’s instructions end with the same limits, whatever you write above them.
Agents look. Nothing changes.
Agents draft. It waits in the Review queue.
None of this happens without Approve.
A published agent gets its own key, limited to reading and to the tools on its list. Publishing a new version replaces the key.
A Workspace can run on Konsil-managed Anthropic, your own key, or another provider.
The Model Context Protocol, an open standard AI assistants use to connect to software. Konsil offers its tools over MCP so your own Claude can use them.
No. It sees what its key allows. You choose the permissions when you create the key.
No. No agent tool sends email or launches a campaign. Campaigns are launched by a person in the console.
No. You answer plain questions. The full wording is there if you want to read it.
Yes. Pause it or archive it from its page.
No. Your data is never used to train models.
We set up each company by hand, so places open a few at a time.
