Overview
Stacklink is the integration, knowledge, and execution boundary for AI agents.
Stacklink connects tools, knowledge, accounts, events, and workbenches for AI agents. It is the backend boundary for external provider integrations and verified knowledge packs, so your app keeps only its own product reasoning and user experience.
The boundary
Stacklink owns external provider accounts, provider toolkits, tool search, tool execution, knowledge packs, verified evidence receipts, triggers, polling, webhooks, jobs, artifacts, and raw provider events. Your consumer app owns final answer generation, product policy, and product-specific UI when using the tool integrations. With Bots, Stacklink also runs the configured model/tool loop and produces the answer. Your application supplies its UI.
What you get
Durable Workflows
Build explicit steps, add a Bot for judgment, and recover saved progress.
Hosted Bots
Configure and publish an agent. Run it through the SDK or your connected AI app.
One session contract
A single session scopes an end-user's tools, connected accounts, and knowledge packs. No per-provider bridge tools.
Agentic tool discovery
Search hundreds of provider tools by intent, fetch exact schemas, then execute — through a small set of meta-tools or over MCP.
Verified knowledge
Citation-backed compliance knowledge with evidence receipts, currentness, and deterministic structured calculations.
Connections & auth
A hosted connect flow for end-users to link provider accounts. Credentials are encrypted and never reach the model.
Triggers & events
Subscribe to provider events via webhook or polling and receive signed, normalized event deliveries.
Remote workbench
A sandboxed runtime where agents run code and bash with provider tools, LLM, and web-search helpers.
The subsystems
| Subsystem | What it does |
|---|---|
| Workflows | Explicit steps, saved progress, triggers and recovery; use visually, through SDK/MCP or embedded in your app. |
| Sessions | Scope tools, accounts, knowledge, and the MCP surface for one tenant/user. |
| Toolkits & tools | ~11 providers, 400+ model-callable tools, discovered by semantic search. |
| Connections & auth | Auth configs (your OAuth app) and connected accounts (the end-user's account). |
| Knowledge packs | Versioned, evidence-backed legal/compliance corpora with structured rule modules. |
| Triggers & webhooks | Provider events in (webhook/polling) and signed events out to your app. |
| Workbench | A secure sandbox where agents run code and bash with provider, LLM, and web helpers. |
| Observability | platform_request_logs and tool_execution_logs for every call. |
Start here
Quickstart
Mount the SDK and run your first tool, in five minutes.
Connect your agent
Drive Stacklink from Cursor, Codex, or Claude over MCP.
Core concepts
The mental model behind the platform.
Reference
Meta-tools, providers, scopes, and configuration.
Start a free trial
The trial includes fixed allowances of tool calls, delivered events and Memory across 2 projects. No card and no automatic charge. Pricing shows the current trial length and allowances; see also Billing & usage.
Asia/Kolkata
The Stacklink runtime is locked to India time. API timestamps are serialized with a +05:30
offset, for example 2026-05-08T05:30:00+05:30.