Changelog - Hexel Studio

June 2026

Complete rewrite of the Hexel Studio documentation.

Accurate API paths, error codes, SDK methods, and enums verified against the codebase
Auth0-style narrative added throughout — plain-English explanations before technical detail
New sections: Data Platform, Orchestration, Tool Gateway, IAM, Observability, Billing
Correct base URL for all API calls (api.hexelstudio.com/compute/v1)
Navigation restructured with product-named tabs and Lucide icons

Hexel Studio is generally available — the infrastructure platform for AI agents in production. Compute

Sandboxes: isolated, single-use execution environments drawn from a warm pool. Allocate in milliseconds, run Python code or shell commands, release when done.
Agent hosting: register a Docker image, deploy it, and receive a permanent endpoint. Same configuration always produces the same URL. Instances run always-on with CPU throttling; idle cost is near zero.
Instances: four compute tiers (micro, standard, large, gpu). Deploy via POST /compute/v1/agents/{id}/instances or hexel compute instance deploy.
Skills: publish reusable capabilities to the platform via /compute/v1/skills.
Sessions and filesystems: group runtime resources and attach persistent storage across sandbox calls.

Data Platform

Knowledge stores: ingest documents (PDF, DOCX, TXT, MD, HTML, JSON, CSV) or connect an S3 datasource. Content is chunked, embedded, and made searchable via semantic search.
Memory: scoped memory stores with automatic expiry — TASK (24h), SESSION (8h), AGENT (1 year), WORKSPACE (1 year).
Connectors: S3 datasource with configurable sync schedule, pause/resume, and sync history.
Context bundles: assemble grounded context from knowledge and memory for a specific run.
All Data Platform APIs at api.hexelstudio.com/data/v1/.

Orchestration

Tasks: submit a goal (input) to a fleet. The platform plans it, executes it across agents, gates sensitive actions with approvals, and streams events over SSE.
Fleets: group agents with runtime configuration (deployment mode, routing policy, capacity profile, isolation profile, egress policy).
Approvals: gate task execution with human or policy review. Pending approvals expire after 24 hours.
Policies: define ALLOW / DENY / REQUIRE_APPROVAL rules evaluated against task context.
Streaming: real-time SSE events (task.submitted, task.state, task.complete, runtime.delta, and more). Full replay from recorded execution.
Orchestration APIs at api.hexelstudio.com/orchestrator/v1/.

Tool Gateway

6 toolkits at launch: Gmail, Google Sheets, Slack, GitHub, Jira, WhatsApp (82 tools total).
Connect accounts via managed OAuth; credentials are encrypted at rest and never returned by the API.
Bind toolkits to agents; the gateway resolves credentials automatically at execution time.
Execution modes: synchronous, asynchronous, batch, preflight. Idempotency keys for safe retries.
20 trigger types across Google Sheets, Gmail, GitHub, Slack, WhatsApp, and Jira for event-driven automation.
Webhook subscriptions with signature verification for execution and trigger notifications.
MCP server support for custom and generated tool integrations.
Tool Gateway APIs at api.hexelstudio.com/tools/v1/.

SDKs & CLI

Python SDK (pip install hexel): compute, orchestration, tools, agent runtime, and LLM instrumentation (OpenAI, Anthropic).
TypeScript SDK (npm install hexel-sdk): compute, orchestration, and tools.
CLI (hexel): infrastructure and operations — compute, fleets, tasks, tools, policies, approvals.
API key and OAuth client credentials authentication via STS (15-minute tokens, 30-day refresh).

Platform

API Reference