HarnessMatch
Menu

Coding agent

Gemini CLI

Google's open terminal agent with sandboxing, subagents, and rollback.

Status
Active
License
Apache-2.0
Evidence
Documented + code-verifiable + independently measured configuration, 17 sources
Product record checked
2026-07-27
Compare harnessesInspect sources

At a glance

An Apache-2.0 terminal agent for Gemini with interactive and non-interactive modes, MCP, policy controls, optional OS or container sandboxing, specialized subagents, worktrees, and file-aware rewind and restore.

Good choice if

  • Organizations retaining Gemini CLI through Gemini Code Assist Standard or Enterprise, Google Cloud, or paid enterprise API access
  • Headless automation with structured output and explicit policy controls
  • Tasks that benefit from subagents, worktrees, sandbox profiles, and file rollback

Check before choosing

  • Since June 18, 2026, Google AI free, Pro, and Ultra consumer accounts are served by Antigravity CLI instead; Gemini CLI remains supported for enterprise and paid API access
  • Model access remains centered on Gemini rather than a multi-provider model catalog
  • OS and container sandboxing is optional; the macOS default profile allows broad reads and network while constraining writes, so profiles must match the task
See 5 more considerations
  • Checkpointing and experimental Auto Memory are off by default; memory candidates require review and may send selected local transcript content to the configured model
  • The browser agent is disabled by default, uses a persistent Chrome profile by default, and changes behavior or availability under Seatbelt and container sandboxes
  • Worktrees isolate checkouts rather than processes, credentials, or network access
  • The policy engine's project-level Workspace tier is currently non-functional, so repository policy files do not enforce tool decisions; use User or Admin policies until that tier is restored
  • The repository contains extensive project-owned eval assets, but no complete independent harness benchmark result is imported

Why it qualifies as a coding harness

This confirms category fit, not product quality. Every required criterion links back to first-party evidence.

Qualifies4 of 4 required criteria evidenced

  • Adaptive agent loop

    Documented

    The system repeatedly observes results and chooses the next action instead of following a fixed one-pass graph.

    EvidenceGemini CLI v0.52.0 source

  • Repository tool execution

    Documented

    The system can use tools to inspect and change a repository or its execution environment.

    EvidenceGemini CLI v0.52.0 source

  • Task-aware context management

    Documented

    The runtime assembles, updates, compacts, retrieves, or persists task-relevant context while work proceeds.

    EvidencePersistent context and memory

  • Model-independent runtime control

    Documented

    Permissions, budgets, interruption, policy, or stop controls operate outside the model's own text generation.

    EvidenceSandboxing

Membership establishes category fit only. It does not score quality, safety, autonomy, model capability, or benchmark performance. · Read the membership rule.

How it works under the hood

Seven mechanisms mapped from first-party records. These labels describe what the harness provides, not how intelligent its model is.

7/7layers documented
  • Execution & isolationSandbox availableDocumented mechanism, not a performance score.
  • Tooling & integrationsExtensible + browserDocumented mechanism, not a performance score.
  • Context & statePersistent stateDocumented mechanism, not a performance score.
  • Lifecycle & recoveryCheckpoint/rewindDocumented mechanism, not a performance score.
  • ObservabilityStructured tracesDocumented mechanism, not a performance score.
  • VerificationTool-assistedDocumented mechanism, not a performance score.
  • Governance & permissionsPolicy controlsDocumented mechanism, not a performance score.

Public code audit

5/5public artifacts present
Security policy
Present at inspected commit
CI workflow
Present at inspected commit
Automated tests
Present at inspected commit
Evaluation assets
Present at inspected commit
Contributor documentation
Present at inspected commit

The v0.52.0 tree contains 963 test-like files, 47 workflows, and 47 eval-like assets spanning behavior, memory, safety, integration, and performance. They are project-owned development evidence without a complete independent model × harness × environment × budget × attempts result, so no product score is imported; hosted Gemini services remain outside the source audit.

Inspect commit d14583b926769bd98f807cdc6b1ca50e91ae26ec, checked 2026-07-27

Capability support

Documented first-class product support, checked against the sources below.

External tools (MCP)
DocumentedProduct-supported MCP integrationSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Local models
Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.
Agent parallelism
DocumentedDelegated or parallel agent workflowSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Runs without an open UI
DocumentedNon-interactive or automation surfaceSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Browser control
DocumentedBuilt-in or product-supported browser controlSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Isolated execution
OptionalOptional OS or container sandbox profilesSource · checked 2026-07-27Profiles and platform support change the effective filesystem and network boundary.
Undo file changes
DocumentedProduct-supported file or session rollbackSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.

Primary evidence

Each capability claim is tied to a first-party record and a verification date.

Product record checked 2026-07-27
Gemini CLI consumer transitionConsumer cutoff date, Antigravity migration, retained enterprise access, API-key access, and ongoing Gemini CLI supportOfficial announcementGemini CLI v0.52.0 sourcePinned stable source, license, tests, workflows, project eval assets, and contributor materialOfficial repositoryStable release 0.52.0Stable version, release date, installation channel, feature highlights, and upgrade guidanceOfficial announcementSandboxingOpt-in Seatbelt, Docker, Podman, gVisor, LXC, Windows isolation, profiles, mounts, and limitsOfficial docsTrusted foldersDefault-off folder trust, untrusted-workspace restrictions, project configuration, MCP, and auto-accept boundariesOfficial docsHeadless modeNon-interactive prompts, text and structured outputs, workspace trust, scripting, and CI behaviorOfficial docsSubagents and browser agentBuilt-in and custom subagents, disabled-by-default browser agent, Chrome modes, consent, policies, and sandbox interactionOfficial docsCheckpointingDefault-off shadow-Git snapshots, conversation and tool-call capture, restore workflow, and requirementsOfficial docs
View 9 additional sources
Additional first-party evidence9 sources
Persistent context and memoryGEMINI.md hierarchy, durable memory files, explicit save, inspection, reload, and context guidanceOfficial docsExperimental Auto MemoryDefault-off transcript mining, review inbox, candidate memory and skills, data flow, privacy, and limitationsOfficial docsMCP integrationMCP server configuration, discovery, transports, trust, authentication, tool filtering, and managementOfficial docsTelemetryDefault-off OpenTelemetry logs, metrics, traces, local and Google Cloud exporters, and data fieldsOfficial docsGemini CLI evaluation assetsProject-owned behavioral and safety evaluations without a complete independent product result recordOfficial repositoryPolicy engineAllow, deny, and ask decisions, rule priority, headless behavior, admin policies, and the disabled Workspace tierOfficial docsLifecycle hooksSession, agent, model, and tool lifecycle hooks, deterministic blocking, configuration scopes, and security implicationsOfficial docsExperimental Git worktreesOpt-in isolated checkouts for sessions and subagents, lifecycle, branch handling, and filesystem-only boundariesOfficial docsEnterprise admin controlsAdministrator-enforced authentication, extensions, MCP, sandboxing, and approval-mode restrictionsOfficial docs