Quick verdict
Our assessment is that OpenCode merits evaluation when model choice matters. Its documented workflows support planning and code changes, while approval settings and provider data handling deserve explicit attention. Version 2 adds web clients and a team Console, keeps plan and build as agents, and does not support session sharing yet.[1][2][3][4][5][6][7][8][9][10]
Capabilities
OpenCode is an MIT-licensed AI coding agent. Version 1 offers a terminal interface, desktop app, or IDE extension; version 2 documents terminal, desktop, and web clients.[1][2][3]
The official documentation lists 75+ LLM providers and local-model support. Authentication can use provider keys, managed-service credentials or supported subscription login; GitHub Copilot is one documented option.[1][2]
Plan proposes changes before Build edits code (modes in version 1, agents in version 2). Undo and redo support revisiting changes, and project initialization creates an AGENTS.md conventions file.[1][2][3]
Permissions use allow, ask, or deny effects. Version 1 groups them by tool with most defaulting to allow; version 2 uses an ordered rules array over an allow-all base policy with ask exceptions, so in either version a command or edit is not guaranteed to receive a separate approval prompt.[1][2]
Who it fits
We would evaluate OpenCode for developers who want a coding agent without committing to one model provider. Local-model support also offers a way to keep inference on infrastructure they control; bringing a hosted-provider key alone does not provide that boundary.[1][2][3]
Organizations can evaluate the documented Enterprise offering for central configuration, SSO and an internal AI gateway. These are documented integration options, not evidence from our own organizational deployment.[1]
Version 2 organizes its team offering as Console, an optional service with model inference, an LLM gateway, usage tracking and budgets, and team-wide policies.[1]
Boundaries to check
Code handling depends on the selected AI provider. In version 1, optional session sharing uploads conversation history and makes it accessible to anyone with the link; version 2 does not support session sharing yet. Neither version should be described as a universal no-storage guarantee.[1][2][3]
The provider docs say Anthropic prohibits the Claude Pro/Max subscription-plugin use described there, and that bundled plugins were removed as of version 1.3.0. Those subscriptions should not be assumed to transfer to OpenCode. Direct provider-key access is documented separately.[1]
One dated Pickuma review reports Claude Code needing fewer turns on the same model before OpenCode context settings were tuned. The tested tool version is unspecified, so this does not establish a current universal performance gap.[1]
Reported experience
One attributed review: Pickuma, displayed publication July 16 and update August 21, 2026, reporting two weeks of work. This refresh reuses that report for community claims; no new community research was conducted.
1 cited community source. Purposive sample of one report, not representative or independently verified. The publisher discloses affiliate links and AI-assisted drafting. Tested tool version and model details are unspecified. Older comparative claims are historical considerations, not current product measurements. Unavailable Aitoolpick, an unverified HN anecdote and the disputed Morph benchmark are omitted.
In a dated review, Pickuma attributes its account to Owen and reports two weeks of use across a TypeScript monorepo and Python service. It describes completed refactoring, test-generation and dependency-upgrade work, alongside the need to tune context and review output. This is the publisher's account, not OpenAlternative testing.[1]
Questions worth asking
Is OpenCode tied to one model provider?
Does every code change require approval?
Can organizations use OpenCode?
Does using OpenCode keep all context local?
What changed in OpenCode version 2?
Sources and editorial review
Source research. Reviewed on October 6, 2026. Written by Sanjib Sah. AI assisted; reviewed by a human editor.
We have not run or tested this product.
One attributed review: Pickuma, displayed publication July 16 and update August 21, 2026, reporting two weeks of work. This refresh reuses that report for community claims; no new community research was conducted. Purposive sample of one report, not representative or independently verified. The publisher discloses affiliate links and AI-assisted drafting. Tested tool version and model details are unspecified. Older comparative claims are historical considerations, not current product measurements. Unavailable Aitoolpick, an unverified HN anecdote and the disputed Morph benchmark are omitted.
- OpenCode - The open source AI coding agent
official · Observed October 6, 2026
Official homepage: open-source AI coding agent for terminal; 75+ providers; Zen/Go managed options; desktop and IDE clients.
- OpenCode docs
official · Observed October 6, 2026
Official docs: install (brew/mise/docker/scoop, anomalyco org), quickstart TUI/IDE setup, Plan/Build modes, checkpoints, sharing, AGENTS.md init.
- GitHub API: anomalyco/opencode repository
official · Observed October 6, 2026
Repository identity anomalyco/opencode, numeric ID 975734319, MIT, public, not archived or a fork. Repository metrics are dated observations, not ratings. Latest-release facts require their separate endpoint.
- OpenCode Review: A Terminal-Native AI Coding Agent
community · Observed October 6, 2026
Pickuma article attributed to Owen; displayed publication July 16 and update August 21, 2026, exact times unknown. Claims two weeks across a TypeScript monorepo and Python service; reports fewer turns with Claude Code before context tuning. Affiliate links and AI-assisted drafting disclosed. Tested product version unknown; independence not verified. Audited October 6, 2026.
- OpenCode docs - Providers
official · Observed October 6, 2026
Official provider documentation supports 75+ LLM providers and local models. Zen uses an OpenCode API key; supported subscription login includes GitHub Copilot. Docs state Claude subscription plugins were unbundled as of 1.3.0 and Anthropic prohibits that use. Re-read October 6, 2026; prior claims unchanged; page carries a version-2 availability banner.
- OpenCode permissions
official · Observed October 6, 2026
Rules allow, ask or deny edits and shell commands. Most permissions default to allow; approval depends on configuration. Reviewed in the October 6 source audit; reused here without claiming a new source read.
- OpenCode Enterprise
official · Observed October 6, 2026
Official organizational offering documents central configuration, SSO and internal AI gateway integration. Code handling is local or through direct AI-provider calls; optional sharing is an explicit exception. Offering is documented, not independently tested. Reused October 6 audit evidence.
- OpenCode session sharing
official · Observed October 6, 2026
Manual sharing is the default and sessions are not automatically shared. Sharing uploads conversation history to the service and creates a public link. Retained until unshared; confidentiality needs this qualification. Reused October 6 audit evidence.
- OpenCode v2 docs
official · Observed October 6, 2026
Version 2 intro: open-source AI coding agent available as terminal-based interface, desktop app, or web app, with install channels and Docker tag 2.0.0.
- OpenCode v2 - Migrate from V1
official · Observed October 6, 2026
Migration guide: three intentional breaking changes (plugin API, server API/clients, terminal config); other supported v1 functionality intended compatible; share policy, ordered permissions, modes to agents, AGENTS.md retained.
- OpenCode v2 - Providers
official · Observed October 6, 2026
V2 providers: models.dev catalog, Copilot device OAuth with Chat access requirement, Ollama/LM Studio/vLLM local discovery; no restated provider count.
- OpenCode v2 - Permissions
official · Observed October 6, 2026
V2 permissions: ordered allow/deny/ask rules over an allow-all base policy with ask exceptions for external directories and .env reads; plan/build shipped as agents; saved approvals are durable project-scoped allow rules.
- OpenCode v2 - Sharing
official · Observed October 6, 2026
V2 sharing page states OpenCode V2 does not support session sharing yet.
- OpenCode v2 - Console
official · Observed October 6, 2026
V2 Console: optional team service with inference, LLM gateway, usage tracking and budgets, team-wide policies; no SSO observed on the intro page.