Quickstart

1. Start a conversation

Create an account — the Free plan needs no card. Open a new chat and pick a model from the model picker: GPT, Claude, Gemini, Grok, Mistral, Perplexity, or any other connected provider.

The part worth trying first: switch models mid-conversation. Ask a question, read the answer, then swap to a different provider and ask a follow-up. The thread carries over — nothing resets. That one interaction is most of what makes multi-model chat different from having three browser tabs open.

2. Add knowledge (optional, but do it)

Create a knowledge base and upload a handful of documents you know well — a policy doc, a product spec, a report you wrote. Akili chunks and indexes them automatically; the indexing state is visible while it works.

Now ask a question those documents answer, with the knowledge base attached. The response cites the passages it used. Check one. That's the evaluation loop this platform is built around — see Knowledge bases for how retrieval works.

3. Create a custom agent

Once a setup works — a prompt, a model, a knowledge base — save it as an agent:

1. Give it instructions: what it is, how it should answer, what it must not do.
2. Pick its model preference and attach the knowledge bases it should ground in.
3. Enable the tools it needs (web search, MCP-backed connectors, code execution on Pro+).
4. Publish it. Every publish is a version — a bad change rolls back in one click.

Agents are the shareable unit: private to you, shared with your org, deployed to Slack, Teams, or your website, or published to the marketplace. The full lifecycle is in Building agents.

4. Understand credits

Every plan includes a monthly credit allowance — a credit is roughly 1,000 tokens of model usage, cost-anchored so heavier models simply draw faster. Usage pauses at zero by default; there are no surprise overage charges. Details in Credits & billing.