Tech Reviews / documentation-only
OpenHands Agent Canvas review: beta and backend scope
AI Summary and verdict
AI Summary: OpenHands Agent Canvas is a documented self-hosted control center for developers coordinating coding agents and automations. This documentation-only assessment preserves its beta status and sandbox caveats; production readiness was not tested. [openhands-readme] [openhands-local-guide] [openhands-sandbox-guide]
Verdict
OpenHands Agent Canvas is a candidate for developers investigating a self-hosted control center across coding agents, backends and automation triggers. Its README labels the project beta and documents distinct filesystem and sandbox boundaries. [openhands-readme] This documentation-based review supports a bounded orchestration trial, not a production-readiness or isolation assurance.
Product facts
- Product
- OpenHands Agent Canvas
- Vendor / project
- OpenHands project
- Category
- coding
- Version / scope
- Official README snapshot checked 2026-10-15; no installed client release tested
- Evidence basis
- documentation-only
- Sources checked
- Verification scope and unknowns
- See the checks performed and unassessed areas
- OpenHands describes Agent Canvas as a self-hosted developer control center for coding agents and automations. (vendor-documented) [openhands-readme]
- The retrieved OpenHands Agent Canvas README labels the project beta. (vendor-documented) [openhands-readme]
- OpenHands documents local execution by default and connections to Docker, VM or company-infrastructure backends for Agent Canvas. (vendor-documented) [openhands-readme]
Capabilities and workflows
Agents and execution backends
- OpenHands documents Agent Canvas use with the OpenHands agent and third-party agents including Claude Code and Codex. (vendor-documented) [openhands-readme]
- OpenHands documents local execution by default and connections to Docker, VM or company-infrastructure backends. (vendor-documented) [openhands-readme]
Automation triggers
- OpenHands documents scheduled and webhook-triggered Agent Canvas automations. (vendor-documented) [openhands-readme]
- For Agent Canvas, a trial should establish where an automation runs and which trigger starts it; a control-center feature does not establish operational reliability. (editorial-opinion) [openhands-readme]
Agent and backend selection
- Agent Canvas documents several coding agents and local, remote or cloud backend choices. An evaluation input is one selected agent/backend pair, a disposable workspace and one reversible task with an explicit expected artifact. (editorial-opinion) [openhands-readme]
- Proposed Agent Canvas evaluation, not executed: identify where the agent server and other services run, select the documented backend, execute the bounded task and inspect the changed files or generated report. Record the agent, backend and granted workspace so the result has a specific scope. (editorial-opinion) [openhands-readme]
- An Agent Canvas success on one backend would not establish parity with other agents or backend configurations. Inspect the actual execution location and access before extending the trial. (editorial-opinion) [openhands-readme]
Scheduled or event-triggered task
- Agent Canvas documents schedule and webhook triggers. Use a harmless report or similar reversible task as the trial input, together with the intended trigger and a place to inspect its output. (editorial-opinion) [openhands-readme]
- Proposed Agent Canvas automation evaluation, not executed: configure the chosen trigger, observe whether a controlled event starts one intended run and inspect its output. The expected evaluation artifacts are the run record and resulting report, not assumed always-on availability. (editorial-opinion) [openhands-readme]
- For Agent Canvas, verify trigger behavior, credentials and workspace access separately; a documented automation feature is not a reliability test or permission guarantee. (editorial-opinion) [openhands-readme]
Isolation boundary before consequential access
- OpenHands warns that the without-a-sandbox path gives the agent server full host filesystem access, and that conversation isolation does not sandbox every service. Those documented boundaries are prerequisites to choosing the evaluation environment. (editorial-opinion) [openhands-local-guide] [openhands-sandbox-guide]
- Proposed Agent Canvas boundary review, not executed: map the services, host paths and workspace grants in the selected setup, then compare that map against your required boundary before granting consequential access. A configuration map is an intended review artifact, not proof of isolation. (editorial-opinion) [openhands-local-guide] [openhands-sandbox-guide]
- Do not infer complete host or service isolation merely because Agent Canvas uses Docker for one part of a conversation. This review did not perform an isolation experiment or security audit. (editorial-opinion) [openhands-local-guide] [openhands-sandbox-guide]
Strengths, limitations, and trade-offs
Strengths
- Agent Canvas documents multiple agent and backend choices, making orchestration scope a concrete evaluation criterion. (editorial-opinion) [openhands-readme]
Limitations and trade-offs
- Agent Canvas’s documented unsandboxed filesystem access and partial isolation boundary require deployment-specific scrutiny; a sandbox label alone is insufficient. (editorial-opinion) [openhands-readme]
Sandbox boundaries matter
- OpenHands warns that its without-a-sandbox option gives the agent server full access to the installation machine’s filesystem. (vendor-documented) [openhands-readme]
- OpenHands states that conversation-level isolation does not place the entire Canvas or automation service inside a sandbox. (vendor-documented) [openhands-readme]
- For Agent Canvas, do not infer that selecting Docker for conversations isolates every service; review the specific documented boundary before granting access to valuable workspaces. (editorial-opinion) [openhands-readme]
Strengths, limitations, and trade-offs
- For Agent Canvas, orchestration flexibility brings deployment responsibility: the place that coordinates agents and the environment that executes a conversation are separate boundaries. The beta label also limits what can be inferred about readiness from the feature list. These are documented scope conditions, not measured availability results. [openhands-readme] [openhands-sandbox-guide] (editorial-opinion) [openhands-readme] [openhands-sandbox-guide]
Best for, not for, and alternatives
Best for
- Consider Agent Canvas when the task is coordinating agent/backend choices and automation triggers, and a developer can own deployment configuration and inspect trial artifacts. This fit follows the documented control-center scope. [openhands-readme]
- Consider an isolated Agent Canvas evaluation when your team can identify the workspace granted to each agent and review the distinction between conversation isolation and the surrounding services. [openhands-sandbox-guide]
Not for / requirements not established
- Agent Canvas’s documented without-a-sandbox option is a mismatch for a policy that prohibits an agent server from having full filesystem access on that host. Evaluate the actual allowed deployment boundary instead of assuming that this option is isolated. [openhands-local-guide]
- Agent Canvas is not established by this review as production-ready or isolated across its services: the source labels it beta and says conversation isolation does not sandbox the whole service. This is a scoped evidence limit, not a claim that every deployment is insecure. [openhands-readme] [openhands-sandbox-guide]
Alternatives
- Evaluate alternatives when a single editor workflow would meet the need, or when your allowed deployment boundary cannot accommodate the Agent Canvas configuration under consideration. Compare orchestration scope, maintained release, workspace access and actual task artifacts; this review has not established a competing control center’s superiority. [openhands-readme] [openhands-sandbox-guide] (editorial-opinion) [openhands-readme] [openhands-sandbox-guide]
Evidence and methodology
Documentation-only review under Tool Review Standard 1.1, content revision r2. Official README excerpts and their specific section locators were rechecked on 2026-10-15; source matching confirms attributed documentation, not independently verified behavior. No product was installed, authenticated or tested; procedures in Part 3 are proposed evaluations. Prior source dates and unresolved conditions are preserved; no numeric score is supported.
Evidence gaps and scope
- Not verified: current commercial prices, account entitlements, provider allowances or total operating cost. No price/value recommendation is made.
- Not assessed: task success, generated-code quality, performance, privacy/data flows, contractual compliance or operational reliability. Those require configuration-specific evidence beyond this documentation review.
- Scope boundary: Agent Canvas is labeled beta. The documented unsandboxed filesystem warning and partial conversation-isolation boundary are not independent security findings.
Disclosure
Sponsored: no. Affiliate links: no.
The publisher confirmed on 2026-10-15 that Global Tech Review independently selected this batch, received no sponsorship or benefits for publication, and uses no affiliate links. Codex assisted with research, writing and automated source review; no product access or hands-on testing was used.
Sources
- OpenHands Agent Canvas — official project README — accessed 2026-10-15
- OpenHands Agent Canvas — Option 1: Without a Sandbox — accessed 2026-10-15
- OpenHands Agent Canvas — Option 3: With Multiple Docker Sandboxes — accessed 2026-10-15
FAQ and further reading
Does Agent Canvas beta status prove it cannot be used?
- No. Agent Canvas’s beta label is a documented lifecycle condition, not an availability or failure test. Assess the consequences for your specific task and support requirements rather than inferring a universal prohibition. (editorial-opinion) [openhands-readme]
Does Agent Canvas conversation isolation sandbox every service?
- No such conclusion follows. OpenHands explicitly says the setting isolates conversation execution without moving the entire Canvas or automation service into a sandbox. Verify the chosen architecture before granting access. (editorial-opinion) [openhands-sandbox-guide]
Does Agent Canvas schedule support establish reliable delivery?
- No. Agent Canvas documents scheduled and webhook-triggered automation, but this review did not measure retries, delivery guarantees, uptime or recovery. Those operational properties remain not assessed. (editorial-opinion) [openhands-readme]