Cliver-project/CLIver: General-purpose AI agent for your terminal
A developer going by the GitHub handle gaol has released CLIver, an open-source AI agent that runs directly inside your terminal. The project brings multi-step, autonomous AI workflows to the command line, which matters because it targets the exact environment where most serious ...
The GitHub repository for CLIver, published under the cliver-project organization and maintained by developer gaol, surfaced on Hacker News in April 2026 with 2 upvotes and zero comments. Low scores on Hacker News do not always predict a project's relevance, and CLIver is worth a closer look than those numbers suggest. The repository has accumulated 190 commits across its history, has been tagged with 5 release versions, and shows a commit as recent as April 17, 2026, meaning this is an actively developed project, not an abandoned weekend experiment.
Why This Matters
Terminal-based AI agents are not a novelty anymore, they are becoming a serious architectural choice for engineering teams that want AI embedded in the same environment where commands run, scripts execute, and infrastructure gets managed. CLIver's feature set, including a CI/CD mode, a workflow engine capable of running LLM-generated directed acyclic graphs, and an OpenAI-compatible API server, puts it in direct competition with tools that have received far more attention and funding. The fact that a solo developer or small team has shipped 190 commits covering parallel task execution, autonomous skill learning, and cost tracking with per-provider currency support in both USD and CNY is a signal that the terminal AI space is moving faster than most coverage of it suggests.
Daily briefing from 50+ sources. Free, 5-minute read.
The Full Story
CLIver is a general-purpose AI agent built to live inside your terminal. The core idea is straightforward: instead of switching to a chat interface or an IDE sidebar every time you need AI assistance, you stay in the terminal and the agent does the heavy lifting from there. It can execute commands, manage workflows, and interact with systems the way a human operator would, but faster and without getting distracted.
The project's commit history tells a more detailed story than the README alone. The most recent major feature push, captured in a commit titled "feat: major feature update," added a substantial list of capabilities. CI/CD mode arrived with a dedicated flag, a JSON output option, configurable timeouts, permission mode controls, and a no-fallback setting for pipelines that cannot afford unexpected model switches. That is not a toy feature. That is something an engineering team would actually use inside a deployment pipeline.
Smart model routing is another standout addition. CLIver can classify errors and automatically fall back to a different model based on capability rather than just availability. It uses jittered exponential backoff for transient errors, which is the kind of production-hardening detail that separates projects built for real use from projects built for demos. When retries are exhausted, it falls to the next model in the routing chain rather than failing silently.
The workflow engine deserves its own paragraph. CLIver can generate a directed acyclic graph of tasks using an LLM, then execute that graph with support for branching, parallel execution, and pause/resume behavior. That puts it in territory usually occupied by dedicated workflow orchestration tools, except CLIver integrates AI generation of the workflow itself. The agent can also learn new skills autonomously by creating reusable skill definitions from complex tasks it encounters, which is a self-improvement loop built directly into the runtime.
On the connectivity side, CLIver includes an OpenAI-compatible API server accessible at the /v1/chat/completions endpoint, messaging platform adapters for Telegram and Discord built on an abstract PlatformAdapter interface, and session management backed by SQLite with full-text search via FTS5. Cost tracking monitors spending per provider and handles the currency difference between providers billing in USD versus CNY, which reflects a genuinely international user base. The project uses Python with the uv package manager for dependency management, a modern toolchain choice that prioritizes fast, reproducible installs.
Key Details
- The repository sits at 6 stars and 0 forks as of the Hacker News submission in April 2026.
- Developer gaol pushed the most recent commit, a uv.lock fix, on April 17, 2026.
- The project has 190 total commits across 1 active branch and 5 tagged release versions.
- CI/CD mode ships with 5 configurable flags: -p, , output json, , timeout, , permission-mode, and , no-fallback.
- Session search uses SQLite FTS5, a full-text search extension built into SQLite.
- The OpenAI-compatible API server exposes the standard /v1/chat/completions endpoint.
- Cost tracking handles both USD and CNY currency denominations across providers.
- Platform adapters support Telegram and Discord through a shared abstract PlatformAdapter class.
What's Next
The project's roadmap, based on the v0.0.4 development branch preparation noted in commits from March 24, 2026, suggests another tagged release is coming soon. Watch for whether the autonomous skill learning feature gets documented more thoroughly, since that capability has the most potential to differentiate CLIver from simpler shell-wrapping AI tools. Community adoption will likely hinge on whether the project publishes a clear quickstart guide, something the AI Agents Daily guides section will be tracking as the project matures.
How This Compares
The most direct comparison is Cline, which released version 2.0 and positioned itself explicitly as turning your terminal into an AI agent control plane, as covered by DevOps.com. Cline has brand recognition, a funded team, and a large community. CLIver has none of those things yet, but it also does not carry the overhead of a commercial product roadmap shaped by investor priorities. The autonomous skill learning and LLM-generated DAG workflow engine in CLIver are features that Cline has not shipped in the same form, which gives CLIver a genuine technical differentiator rather than just a positioning one.
InfoQ's coverage of agentic terminals, published under the heading "How Your Terminal Comes Alive with CLI Agents," confirms that enterprise architecture communities are now treating terminal AI agents as legitimate infrastructure components. That framing matters because it expands the potential audience for a tool like CLIver beyond solo developers and into DevOps teams and platform engineering groups. A tool with an OpenAI-compatible API server and a CI/CD mode is not pitching itself to hobbyists.
The Medium roundup "10 Must-Have CLIs for Your AI Agents in 2026" by Unicodeveloper signals that the space is already crowded enough to generate listicle coverage. CLIver is almost certainly not on that list yet, because its Hacker News visibility is minimal. But the feature set described in its commit history is competitive with tools that have made those lists. The window for CLIver to build a community before the space consolidates around 2 or 3 dominant tools is probably measured in months, not years. You can track related AI tools in this category as the field narrows.
FAQ
Q: What is CLIver and what does it actually do? A: CLIver is an open-source AI agent that runs inside your terminal. You interact with it from the command line, and it can execute commands, build and run multi-step workflows, manage sessions, and connect to AI model providers. Think of it as having an AI assistant that lives where your work already happens, rather than in a separate chat window.
Q: Do I need to pay for AI models to use CLIver? A: CLIver connects to external AI model providers, which typically charge for API usage. The project includes cost tracking that monitors spending per provider and supports both USD and CNY currency denominations, so you can watch what you are spending. The CLIver software itself is open source and free to use.
Q: How is CLIver different from just typing prompts into a chatbot? A: CLIver operates as an agent, meaning it can plan and execute sequences of steps rather than responding to a single prompt and stopping. Its workflow engine can generate a directed acyclic graph of tasks using an LLM and then execute those tasks in parallel or sequence, with branching logic and pause/resume support, which is far beyond what a chatbot conversation can . CLIver is the kind of project that flies under the radar precisely because it is built for people who do not need it explained to them. The engineering is solid, the feature set is ambitious, and the open-source model gives it room to grow without a product team making decisions for users. Subscribe to the AI Agents Daily weekly newsletter for daily updates on AI agents, tools, and automation.
Get stories like this daily
Free briefing. Curated from 50+ sources. 5-minute read every morning.
