It sounds like a migration.
Teams stall because adding trust seems to mean re-platforming the agents they already run.
Connect the agents you already run, and everything they should know.
Point your agents at AISquare with one URL and one header, or wrap them with the SDK. Connect Gmail, Drive and GitHub so every agent works from what your company knows. Nothing to rip out.
Agents in · proxy or SDK
Claude Code
Codex
LangChain
Agno
n8n
OpenAI
Anthropic
Gemini
AISquare
Trace · Reason · Govern · Learn
Knowledge in · connectors
Gmail
Google Calendar
Google Drive
OneDrive
Dropbox
GitHub
The problem
Trust shouldn't mean a rebuild.
One tool per model, per framework.
Every new stack brings its own monitoring, and nothing is shared across them.
Your knowledge sits where agents can't see it.
Policies, playbooks and past decisions live in mail, drives and repos your agents never read.
Connect your agents
Three ways in. Pick per team.
Every run lands in the same reasoning graph, however it got there, so teams can mix paths without redoing anything.
Proxy, zero code.
Point your provider base URL at AISquare and add one header. Your provider key passes straight through to the model.
- Setup in about 5 minutes
- No new dependency, no agent code change
- Anthropic, OpenAI, Azure OpenAI and Gemini
- Rules enforced at the model call, on our side
agent.py
from anthropic import Anthropic
client = Anthropic(
base_url="https://<proxy-host>",
default_headers={
"X-AISquare-Key": "AIS_<your-key>",
"X-Agent-Name": "support-bot", # optional, names the agent
},
) # your ANTHROPIC_API_KEY passes throughNot sure which path fits? Compare them in the docs
Connect your knowledge
Your company's knowledge, one click away.
Connect the tools where your knowledge already lives. Sources feed the Collective Intelligence Graph, and each Studio picks the ones it needs.
- Sign in once, no exports
- Search or describe what you want to connect
- Every answer cites the source it came from
Browse all connectors
Search connectors or describe what you want to connect...
Mail & calendar
Gmail
Google Calendar
Outlook
Files & storage
Google DriveConnected
OneDrive
Dropbox
Docs & wikis
Notion
Confluence
Chat & meetings
SlackConnected
Microsoft Teams
Zoom
Don't see yours? Connect any MCP server or talk to us.
Integrations
Everything it connects to.
Anthropic
Model providers
OpenAI
Model providers
Azure OpenAI
Model providers
Google Gemini
Model providers
LangChain
Agent frameworks
Agno
Agent frameworks
n8n
Agent frameworks
Custom agents
Agent frameworks
OpenTelemetry
Agent frameworks
Claude Code
Coding agents
Codex
Coding agents
Gmail
Knowledge sources
Google Calendar
Knowledge sources
Google Drive
Knowledge sources
Dropbox
Knowledge sources
OneDrive
Knowledge sources
GitHub
Knowledge sources
Built on OpenTelemetry, so the spans you already emit come in as-is. Don't see yours? Talk to us.
Rollout
Start in Audit. Go Live when you're ready.
Connect in Audit
Traces, reasoning, policy verdicts and cost on every agent. Nothing is altered at runtime.
Turn on the rules that matter
Switch the rule books that matter to Live for agents that handle sensitive data or take irreversible actions.
Bring the fleet on
Each team picks the path it prefers. Every run lands in the same place, so nothing is redone.
Your keys stay yours
Provider keys pass straight through and are never stored.
Reversible
Disconnect any time by removing the base URL or running off.
Fails closed
If the gateway is down, agents that enforce rules are never silently ungoverned.
Sensitive data screened
PII, PHI and PCI are detected before anything is saved, and can be masked.
FAQ
Questions about connecting AI agents and knowledge sources.
Proxy or SDK, which should I use?
Start with the proxy for zero-code observability and enforcement across your fleet. Add the SDK to agents where you want the policy gate inside the agent, at the tool boundary, or custom spans. You can mix both, and every run lands in the same reasoning graph.
Does the proxy cover tool calls?
Yes. The proxy enforces at the model call, on AISquare's side. With a rule book in Live it can block an unsafe tool call and hand the agent the reason, rewrite a non-compliant reply, flag a run, or check tool outputs for prompt injection before the model sees them.
Which models and frameworks are supported?
The proxy supports Anthropic, OpenAI (Chat Completions and Responses), Azure OpenAI and Gemini, plus the Claude Code and Codex CLIs. The SDK supports Agno and LangChain with automatic tool interception, custom agents and OpenAI SDK callers through GovernedAgent.wrap, and existing OpenTelemetry setups.
Do I need to change my agent code?
Not with the proxy. You change the base URL and add one header. The SDK needs one package and a wrapper around your agent.
Does our data leave our environment?
Your provider keys pass through to the model and are never stored. The only added destination is your own AISquare workspace, over TLS. PII, PHI and PCI are screened at ingest and can be masked or blocked per workspace.
Can I try this without affecting production?
Yes. Start with rule books in Audit. AISquare records and flags runs but never changes them. Switch to Live only when you're ready.
Which knowledge sources can I connect?
Gmail, Google Calendar, Google Drive, Dropbox, OneDrive and GitHub today. Sources feed the Collective Intelligence Graph, and each Studio chooses the sources it uses.
We already use OpenTelemetry. Do we re-instrument?
No. The SDK attaches to your existing TracerProvider and captures the spans you already emit. You add an agent name so each run routes to the right Studio and rule book.
Add a trust layer without a rewrite.
Start your journey with AISquare
Connect the agents you already run and see every run's reasoning in days, not quarters.