> This is the markdown version of https://www.maniac.ai/integrations. Visit the full page for interactive content. \[ Integrations \] # One endpoint. A thousand tools. Your agent talks to one tool. Arendil talks to all of them — securely, with delegated auth, and with a pull-request workflow on every write action. [Start with one connector](https://app.maniac.ai/auth/register)[Book a walkthrough](/book-demo) 1,000+ tools, 80+ categoriesSSO · SOC 2 · BYO modelReviews on every write agent.py arendil v0.4 ``` from arendil import ToolAgent # One tool exposed to your model. # 1,000+ underlying tools available — none in context. agent = ToolAgent( user="sarah.chen@acme.com", review_policy="auto_under_$1k", ) result = agent.ask( "Mark opp 0064x as Closed Won, $148k ARR," " then notify the deal team in #revenue." ) # → SALESFORCE_UPDATE_OPPORTUNITY ⏸ awaiting review # → SLACK_POST_MESSAGE ✓ auto-approved ``` drop-in replacement for your tools dict zero schemas in your context \[ The catalog \] ## 1,000+ tools, behind one agent endpoint. Add a connector once. Every agent in your org gets it — with auth, retries, idempotency, and review state already wired up. ![Salesforce logo](https://logos.composio.dev/api/salesforce)Salesforce ![Hubspot logo](https://logos.composio.dev/api/hubspot)Hubspot ![Slack logo](https://logos.composio.dev/api/slack)Slack ![GitHub logo](https://logos.composio.dev/api/github)GitHub ![Linear logo](https://logos.composio.dev/api/linear)Linear ![Jira logo](https://logos.composio.dev/api/jira)Jira ![Notion logo](https://logos.composio.dev/api/notion)Notion ![Stripe logo](https://logos.composio.dev/api/stripe)Stripe ![Snowflake logo](https://logos.composio.dev/api/snowflake)Snowflake ![Shopify logo](https://logos.composio.dev/api/shopify)Shopify ![Asana logo](https://logos.composio.dev/api/asana)Asana ![Zendesk logo](https://logos.composio.dev/api/zendesk)Zendesk ![Intercom logo](https://logos.composio.dev/api/intercom)Intercom ![Gmail logo](https://logos.composio.dev/api/gmail)Gmail ![Google Calendar logo](https://logos.composio.dev/api/googlecalendar)Google Calendar ![Google Drive logo](https://logos.composio.dev/api/googledrive)Google Drive ![Figma logo](https://logos.composio.dev/api/figma)Figma ![Datadog logo](https://logos.composio.dev/api/datadog)Datadog ![Pagerduty logo](https://logos.composio.dev/api/pagerduty)Pagerduty ![Segment logo](https://logos.composio.dev/api/segment)Segment ![Zoom logo](https://logos.composio.dev/api/zoom)Zoom ![Microsoft teams logo](https://logos.composio.dev/api/microsoft_teams)Microsoft teams ![Google Sheets logo](https://logos.composio.dev/api/googlesheets)Google Sheets ![Bitbucket logo](https://logos.composio.dev/api/bitbucket)Bitbucket ![Airtable logo](https://logos.composio.dev/api/airtable)Airtable ![Mailchimp logo](https://logos.composio.dev/api/mailchimp)Mailchimp ![Sendgrid logo](https://logos.composio.dev/api/sendgrid)Sendgrid ![Gitlab logo](https://logos.composio.dev/api/gitlab)Gitlab ![Databricks logo](https://logos.composio.dev/api/databricks)Databricks ![Discord logo](https://logos.composio.dev/api/discord)Discord ![Dropbox logo](https://logos.composio.dev/api/dropbox)Dropbox ![Monday logo](https://logos.composio.dev/api/monday)Monday ![Pipedrive logo](https://logos.composio.dev/api/pipedrive)Pipedrive ![Mixpanel logo](https://logos.composio.dev/api/mixpanel)Mixpanel ![Amplitude logo](https://logos.composio.dev/api/amplitude)Amplitude ![Metabase logo](https://logos.composio.dev/api/metabase)Metabase ![Sentry logo](https://logos.composio.dev/api/sentry)Sentry ![Vercel logo](https://logos.composio.dev/api/vercel)Vercel ![Cloudflare logo](https://logos.composio.dev/api/cloudflare)Cloudflare ![Posthog logo](https://logos.composio.dev/api/posthog)Posthog ![Workday logo](https://logos.composio.dev/api/workday)Workday ![Docusign logo](https://logos.composio.dev/api/docusign)Docusign ![Calendly logo](https://logos.composio.dev/api/calendly)Calendly ![Clickup logo](https://logos.composio.dev/api/clickup)Clickup ![Google BigQuery logo](https://logos.composio.dev/api/googlebigquery)Google BigQuery ![ServiceNow logo](https://logos.composio.dev/api/servicenow)ServiceNow ![Freshdesk logo](https://logos.composio.dev/api/freshdesk)Freshdesk ![Quickbooks logo](https://logos.composio.dev/api/quickbooks)Quickbooks 01 · proxy Just-in-time tool selection. Your model only sees one tool: ask\_the\_tool\_agent. A sub-agent owns the full registry and picks the right call from intent. Add tools without re-prompting. 02 · ci/cd Pull requests for every write. Reads run free. Writes are intercepted, queued, and reviewed — by humans, by policy agents, or both. Approve, reject, comment, replay, rollback. 03 · auth Delegated auth, sealed secrets. OAuth, API keys, and SSO live in the connector plane. Your agent never sees a token — and can never leak one. Per-user scopes inherited from your IdP. \[ Watch the round-trip \] ## One ask. Two actions. One review. The proxy agent decomposes intent into concrete tool calls. The gate decides which ones need a human. The connector plane runs what’s approved. Your agent stays out of the weeds. Intent in Free-form natural language from your agent. Calls out Concrete, typed tool calls with idempotency keys. Reviews where it matters Per-tool, per-action policies — humans only on the high-stakes ones. Audit + replay Every call, every approval, every rollback — stored, searchable. arendil run · trace\_id 9f2c…7b1 3.4s · 2 calls · 1 review 1. sarah →Mark opp 0064x as Closed Won, $148k ARR. Notify the deal team. 2. agent →ask\_the\_tool\_agent("close opp 0064x; ping #revenue") 3. proxy →resolve intent → SALESFORCE\_UPDATE\_OPPORTUNITY 4. proxy →resolve intent → SLACK\_POST\_MESSAGE 5. gate ⏸PR #482 · SALESFORCE\_UPDATE\_OPPORTUNITY · awaiting human reviewer 6. gate ✓SLACK\_POST\_MESSAGE · auto-approved (read-mostly · within budget) 7. review →marcus.oyelowo@ approved PR #482 (1m 42s) 8. exec ✓salesforce.update\_opportunity(0064x, stage="Closed Won", arr=148000) 9. exec ✓slack.post\_message(channel="#revenue", text="🎉 Closed Won — Acme · $148k ARR") 10. agent ←done. 2 actions, 1 review, 0 schemas in context. every line replayable, every approval auditable done \[ Browse by category \] ## Pick the one you need today. Add the rest as you go. Every tile below is one connector away. Wire it once, make it available to every agent in your org. [search the full catalog](#) 120+ CRM & sales Salesforce, HubSpot, Pipedrive, and the long tail. Salesforce Hubspot Pipedrive 90+ Communication Slack, Discord, Teams, Zoom, every inbox you can think of. Slack Discord Microsoft teams Gmail 180+ Data & analytics Warehouses, BI, event streams, and reverse-ETL. Snowflake Segment Mixpanel Amplitude 210+ Developer tools Source, observability, deploys, schedulers. GitHub Sentry Datadog Vercel 150+ Productivity Notion, Asana, Linear, Jira, ClickUp, and friends. Notion Linear Jira Asana 110+ Finance & ops Stripe, Shopify, NetSuite, Workday, the back office. Stripe Shopify Workday Docusign Don’t see the tool you need? We add new connectors weekly, and ship custom ones for enterprise customers. [Request a connector](/contact) \[ Reviews for agent writes \] ## Every side effect is a pull request. Approve, reject, replay. Reads run at agent speed. Writes pause at a review queue you control — humans for the high-stakes ones, policy agents for the boring ones, full audit trail either way. review\_queue · open · last 60 minutes LIVE 1. PR #482rev-ops-agent salesforce.update\_opportunity set Stage = "Closed Won" on Opp 0064x00000abc ``` - Stage: "Negotiation" + Stage: "Closed Won" + ARR: $148,000 ``` reviewer · human · marcus.oyelowo@ awaiting review 2. PR #481support-triage zendesk.update\_ticket reassign 142 stale P3 tickets to overflow queue ``` - assignee: original_owner + assignee: overflow_queue + tag: "auto-rerouted" ``` reviewer · agent · policy-bot · within budget auto-approved · policy 3. PR #480finance-agent stripe.create\_refund refund $2,400 to cus\_NbX… (duplicate charge) ``` + amount: 2400.00 USD + reason: "duplicate" + idempotency_key: … ``` reviewer · human · sarah.chen@ · cfo office approved 4. PR #479data-cleanup postgres.delete DELETE 38,402 rows from prod.customer\_events ``` - WHERE event_ts < now() - interval '30 days' - estimated cascade: 4 tables ``` reviewer · human · platform-on-call rejected Define policies per-tool: auto-approve under $1k, escalate deletes, require two humans on production schema changes. The rest is just a queue. POLICIESREPLAYROLLBACK \[ Auth + security \] ## Built so your security team approves the deploy. Same controls whether you run in our cloud or inside your VPC. SSO, SCIM, SOC 2, and an audit trail that maps cleanly to your existing review workflows. Delegated OAuth Each user authenticates once. The connector plane stores tokens. Your agent passes a user ID — never a credential. Vault for secrets API keys and service credentials live encrypted at rest, scoped per workspace, rotated on demand. Never in a prompt. Sandboxed execution Connector code runs in isolated workers with per-call IAM. A misbehaving agent can’t exfiltrate sibling integrations. Policy as code Per-tool, per-action, per-environment rules. Tighten as your agents earn trust. Loosen when it’s safe. BYO model OpenAI, Anthropic, open-source, or your own endpoint. We never lock the reasoning layer. Full audit trail Every call, every approval, every rollback — stored with the user, the trace, and the diff. Exportable to your SIEM. \[ The context-window math \] ## Stop paying for tool schemas you don’t use. On every turn. The tool-calling sub-agent owns the schema registry and picks the right call from intent. Your agent sees one tool, regardless of how many connectors you wire up. WITHOUT THE PROXY Schema bloat in every turn. - 300 tool schemas pasted into every prompt - Token budget burned before reasoning starts - Every new connector regresses behavior - Secrets and auth flow through the model - No way to gate writes without rewriting prompts WITH THE PROXY One tool. Hundreds available. Zero overhead. - One tool: ask\_the\_tool\_agent(intent) - Sub-agent owns the registry and routing - Add connectors without touching your agent - Secrets stay in the connector plane - Writes are gated by policy, not prompt-engineering secrets sealed in connector plane·add tools without re-prompting \[ Next step \] ## Wire one tool. Ship one approval. Connect a single system, point your agent at Arendil, and watch the first write get gated, reviewed, and replayed. No prompt rewrites, no secrets in your model, no schemas in your context. [Start with one connector](https://app.maniac.ai/auth/register)[Book a technical walkthrough](/book-demo)[Talk to our enterprise team](/contact) policies tighten as your agents earn trustSOC 2 · VPC · BYO modelevery write reviewable, replayable, reversible --- *Arendil, High throughput background agents. Opus-quality outputs at 1/50 of the cost. Learn more at [maniac.ai](https://www.maniac.ai).*