Dev & Engineering coding-clicode-generationdebuggingcommand-executionfile-operationsmulti-providerweb-dashboardmodel-fine-tuning

Supercode

A development platform combining a terminal coding agent, web dashboard, and multiple model providers.

FollowAgents review · FARS-2.1
Not recommended
46/ 100 5-point scale 2.3 / 5
1 2 3 4 5 6
Per-dimension scores and reasoning
1Trust10 / 29 · 1.7/5

The evidence shows explicitly scoped GitHub permissions in the desktop workflow, a temporary signing keychain, disabled command tracing, cleanup of signing material, and clear repository/license attribution. This supports stronger sensitive-data handling and source attribution scores. Deductions apply because these controls primarily cover release automation, while the Agent is said to support file writes, command execution, multiple model providers, and external databases without demonstrated sandboxing, path or command restrictions, per-action confirmation, complete data-flow disclosure, operation rollback, or Agent-specific dependency auditing and vulnerability response. No red-line behavior is evidenced.

2Reliability8 / 14 · 2.9/5

The README, root package manifest, and CI are broadly consistent about Bun, Turborepo, major applications, and quality commands. Bun 1.2.21 is pinned, and CI installs from the frozen lockfile before type checking, linting, and testing, providing reasonable dependency-availability evidence. Deductions apply because some documented script names do not appear in the supplied root manifest, including db:migrate and db:terminal:migrate; database generation is allowed to fail before tests; and beyond a claimed graceful Prisma skip when DATABASE_URL is absent, no concrete failure messages are shown for Agent tools, providers, or network errors.

3Adaptability8 / 18 · 2.2/5

The material clearly targets software developers and names code generation, debugging, explanation, documentation, and multilingual scenarios. It describes local dashboard, CLI, terminal client, multiple model providers, and the Bun, Docker, PostgreSQL, and environment requirements. Deductions apply because the Agent's permission boundaries, supported and unsupported tasks, tool-trigger rules, confirmation conditions, and provider differences are undefined. There is no static evidence for trigger precision, and environment fit is documented mainly for local development.

4Convention10 / 18 · 2.8/5

The README thoroughly organizes architecture, directories, applications, packages, scripts, and environment setup. The complete MIT text matches the license metadata, justifying full marks for information architecture and licensing. Installation guidance and examples are useful, but CLI-specific environment setup is deferred to an unavailable subdirectory and no FAQ is supplied. Supercode, Supercli, and super-platform coexist as names, while a desktop application present in the workflow is absent from the documented tree. Only limited status information is given; there is no comprehensive limitations section, changelog, or clear Agent maintenance and release-responsibility path.

5Effectiveness6 / 13 · 2.3/5

The combination of a coding Agent, dashboard, terminal client, provider flexibility, and shared tools offers plausible marginal value over a basic single-purpose CLI, and the README supplies actionable launch commands. Deductions apply because no representative Agent outputs, completed interactions, quality measurements, or recovery experience are shown, so output usability is only thinly supported. Operation may also require model APIs, databases, Docker, Pinecone, or other services, while the only cost discussion concerns model training rather than routine Agent usage and benefit tradeoffs.

6Verifiability4 / 8 · 2.5/5

Core product, runtime, and script claims can often be checked across the README, package.json, and CI. A pinned runtime, frozen-lockfile CI, and concrete workflow steps improve traceability. Deductions apply because the supplied evidence omits the Agent's tool implementations, configuration, tests, lockfile, and CLI-specific documentation. Claims about sponsorship, publication, capabilities, and graceful installation behavior lack corresponding implementation evidence, and roadmap goals, current capabilities, and promotional language are not consistently separated.

Evidence confidence: Low Reviewed Sep 11, 2026 Reviewed revision fbc5af280127
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Safety controls not found in source: confirmation before acting, rollback or recovery path
Before you use it
  • The Agent is described as reading and writing files and executing commands, but the supplied material does not establish sandboxing, command or path restrictions, dangerous-action confirmation, or rollback. Review those implementations in an isolated environment before granting repository or terminal access.
  • Multiple model providers, OAuth, databases, a vector store, and background jobs may transmit source code, prompts, identity data, or telemetry to third parties. Do not process sensitive code or credentials until complete data flows and retention policies are established.
  • Some installation and development commands differ from the root scripts, and CI permits database generation to fail. Manually verify actual scripts, the lockfile, and CLI environment variables at the pinned revision before deployment.
  • The release-signing workflow handles secrets carefully, but that does not substitute for auditing runtime dependencies, Agent tool permissions, and supply-chain vulnerabilities.
Review evidence [1][2][3][4][5]
See the full review method →

What does this agent do, and when should you use it?

Supercode is a full-stack software development platform organized as a Bun and Turborepo monorepo, with an npm-distributed `supercode` coding-agent CLI at its center. The repository also contains a Next.js management dashboard, an MDX documentation site, a browser terminal client, backend scaffolding, and shared packages. Its CLI uses AI SDK v6 with OpenRouter, Anthropic, and Google integrations and exposes file reading, file writing, command execution, and search tools. The dashboard handles repository management, analytics, and GitHub OAuth, while the dashboard and terminal products have separate Prisma database schemas. A parallel open-model project trains Qwen3-8B and GLM-4-9B with LoRA or QLoRA, although that work is still described as being in training and evaluation. The project best fits teams prepared to operate a Bun monorepo and configure their own model credentials, authentication, and database infrastructure.

A user starts the supercode CLI from apps/supercode-cli/server and interacts with its terminal chat loop. The agent can call its tool implementations to read and write files, execute commands, and search, while the service layer reaches supported models through AI SDK v6, OpenRouter, Anthropic Claude, and Google. The Next.js application in apps/supercode-cli/client provides a browser interface modeled on the terminal experience. Separately, apps/web manages repositories, displays analytics, and implements GitHub OAuth using shared database, authentication, UI, and background-job packages. Packages such as @super/claude-sdk, @super/embeddings-sdk, and @super/skills expose reusable integration code. The supercode-openmodel project prepares coding datasets, trains Qwen3-8B and GLM-4-9B variants, evaluates them, and is intended eventually to supply the CLI with the best-performing model.

  1. A developer who wants a conversational terminal agent that can generate or explain code, edit project files, and run commands.
  2. An engineering team that wants to configure coding workflows around Anthropic, OpenRouter, and Google-backed capabilities rather than a single documented provider.
  3. An internal-tools team that needs both an installable CLI and a browser-based terminal interface for the same product.
  4. A team that wants a web dashboard for repository management and analytics with GitHub OAuth authentication.
  5. An ML engineering group experimenting with LoRA or QLoRA fine-tuning of Qwen3-8B or GLM-4-9B for coding tasks.

What are this agent's strengths and limitations?

Pros
  • One monorepo supplies an installable CLI, browser terminal, management dashboard, documentation site, and reusable SDK packages.
  • The CLI explicitly supports OpenRouter, Anthropic, and Google integrations through a shared AI SDK service layer.
  • Its file read/write, command execution, and search tools connect model output to concrete repository operations.
  • Most CLI operations do not require a database, allowing evaluation of the terminal agent before deploying the full dashboard.
  • The project exposes test, ESLint, TypeScript-checking, and separate database migration commands for its major components.
Limitations
  • A complete deployment requires Bun, PostgreSQL, environment configuration, and several external services, making it heavier than a standalone CLI.
  • The dashboard requires database settings, a Better Auth secret, and GitHub OAuth credentials before its principal features can operate.
  • The dashboard and terminal maintain separate Prisma schemas and migration commands, adding operational overhead.
  • Core AI features depend on networked model providers, while the source provides no pricing, quota, or offline-inference guarantees.
  • The open-model effort remains in training and evaluation, with no reported benchmark results or production-ready checkpoint.
  • apps/api is described only as scaffolded and its port remains TBD, so it should not be treated as a finished shared backend.

How do you install or deploy this agent?

Prerequisites are Git and Bun 1.2+; the repository pins Bun 1.2.21. The dashboard also needs PostgreSQL, commonly through Docker Desktop, or a remote PostgreSQL provider such as Neon or Supabase. Run:

git clone https://github.com/yashdev9274/supercli.git
cd supercli
bun install
cp apps/web/.env.example apps/web/.env.local

Configure at least DATABASE_URL, BETTER_AUTH_SECRET, BETTER_AUTH_URL, GITHUB_CLIENT_ID, and GITHUB_CLIENT_SECRET in apps/web/.env.local. For a local database, continue with:

docker compose up -d
bun run db:migrate
bun run dev:web

The dashboard is then available at http://localhost:3000. Most CLI operations can be used without setting up a database.

How do you use this agent?

From the repository root, run bun run supercode for the first CLI session in development mode. To build and use the production output, run bun run build && bun run supercode:prod. Start the browser terminal with bun run dev:terminal, the CLI development loop with bun run dev:terminal-server, or the documentation site with bun run dev:docs. bun run dev starts all Turborepo development services. AI functionality requires credentials for the selected provider; the supplied source does not enumerate the CLI's complete environment-variable names, so those must be confirmed in apps/supercode-cli/ configuration.

FAQ

Does the CLI require PostgreSQL?
Not for most operations. Installation skips Prisma client generation when DATABASE_URL is absent, although the dashboard and database-backed features do require PostgreSQL.
Which model providers are documented?
The CLI names OpenRouter, Anthropic, and Google integrations, with Anthropic Claude identified as the default model. Google is also used for embeddings. Relevant provider credentials are required.
What local permissions does the agent need?
Its tools include file reads and writes, command execution, and search. The process therefore needs filesystem and shell access appropriate to the workspace, and adopters should constrain those permissions through their execution environment.
Can the terminal agent run without the web dashboard?
Yes. bun run supercode launches the CLI independently, while the dashboard, documentation site, and browser terminal have separate commands.
Is the repository's open coding model production-ready?
The source does not establish that. It is explicitly marked as in progress, with training and evaluation underway before a best-performing model is selected for CLI deployment.

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents