ChatCLI - ChatCLI
Skip to main content

Your terminal becomes an AI engineer.
Fourteen providers, coder mode, agents, a gRPC server and a Kubernetes operator: one Go binary.

New releasev1.223.0move the whole state directory with CHATCLI_STATE_DIR; date flags and dated hits for session and memory recall; pin the date ChatCLI reasons about with CHATCLI_NOWRead release notes →

Quickstart

Install ChatCLI, connect a provider and chat in under five minutes.

Coder mode

reads, edits, runs tests and rolls back, in a loop you approve.

Agents and harness

Multi-agent squads, task graphs and seven reliability patterns.

Server and Kubernetes

Share providers with a team over gRPC and run AIOps in your clusters.

​
Quick start

1

Install ChatCLI

Every install path, including Helm, building from source and OAuth logins, is on the Installation page.
2

Configure a provider

Each provider has a default model; set <PROVIDER>_MODEL to pick another, or use /switch at runtime. Have a ChatGPT, Claude or Copilot plan? Skip the key and run /auth login <provider> inside ChatCLI. All fourteen providers are in Supported models.
3

Start chatting

You are talking to an LLM in your terminal, with your project’s context one @ away. Next, try coder mode.

​
Connect a channel

Use ChatCLI from your chat app. Every message runs the full agent, with its tools and your memory.
TelegramSlackDiscordWhatsAppWebhookHow the gateway works

​
See it in action

One prompt, real work: /coder runs the failing tests, reads the code, patches the bug and re-runs go test uncached to prove the fix — live, on Claude Sonnet 5.5

In practice, your everyday terminal turns into this:

​
Memory, measured

On LongMemEval-S — 500 questions, each over about 115k tokens of past conversations — chatcli answered 90.8% by searching its own saved sessions, and 90.0% on the questions never used to tune it. Its search surfaced the evidence in 469 of 470 answerable questions.

See the full LongMemEval results

The method, every number, where it fails, what it changed in chatcli, and the data to rerun it.

​
Browse the docs

Get started

Quickstart, installation, Docker and Kubernetes, updates.

Using ChatCLI

The prompt, @ context, agent and coder modes, one-shot runs.

Coder

The @coder tool, atomic tools, permissions, worktrees, LSP.

Agents

Multi-agent, squads, task graphs, personas and the quality harness.

Context & memory

Persistent context, knowledge bases, memory, sessions, token efficiency.

Models & providers

Fourteen providers, OAuth, Bedrock, fallback, rate limits and cost.

Tools

Web search, browser, API explorer, Git forges, images, diagrams, scheduler.

Extensions

Plugins, MCP servers, skills, slash commands and hooks.

Channels & gateway

ChatCLI on Telegram, Slack, Discord and WhatsApp, with voice replies.

Server & IDE

The gRPC server, remote connections, MCP server and ACP for editors.

Kubernetes & AIOps

K8s watcher, the AIOps operator, incidents, SLOs and remediation.

Security

Enterprise security controls and dependency vulnerability scans.

Cookbook

Step-by-step recipes for refactoring, debugging, pipelines and AIOps.

Reference

Commands, environment variables, configuration and the REST API.

Releases

Release notes for every version, with links to the changes.

​
What is ChatCLI?

ChatCLI brings the leading large language models (LLMs) straight into your shell. It understands your project’s context, edits code, runs commands and automates complex tasks, without you ever leaving the command line. A single Go binary: fast, portable and self-contained.

14 providers

OpenAI, Claude, Gemini, Grok, Ollama and more — in a single interface.

350+ models

One unified catalog. Switch model at any time.

37+ native tools

@coder, @lsp, @proc, @http, @api-explorer, @knowledge, @memory and many more.

7 quality patterns

ReAct · Reflexion · CoVe · RAG+HyDE · Self-Refine…
The CLI is the main product — it works 100% locally, no server required. The server and operator are optional for team and Kubernetes scenarios. All versions are published automatically on every release via GitHub Actions. Check the full history on Releases or ArtifactHub.
Who is it for?
  • Developers: debug code, understand codebases, generate tests, refactor and document, all without leaving the terminal.
  • DevOps and SREs: analyze K8s logs, automate deploys, diagnose incidents with AI and monitor clusters.
  • CLI enthusiasts: supercharge your terminal, create powerful aliases and explore new ways to work with AI.
  • DBAs and data engineers: automate repetitive tasks, analyze queries, manage databases and data pipelines.
What does it replace?
  • Endless copy and paste: no more cat file.js, selecting, copying, opening the browser and pasting. Use @file right in the terminal.
  • Generic commits: @git plus a question gives a commit message based on the actual diff.
  • Intimidating log analysis: pipe straight to the AI with cat error.log | chatcli -p "root cause?".
  • A steep learning curve: @file ./src --mode smart and ask “how does auth work?”, and onboarding takes minutes.

​
How it works

ChatCLI ships as three artifacts that work standalone or together: the local CLI, the gRPC server, and the Kubernetes operator. Use just the CLI on your terminal, or set up a full topology with server + operator for teams and production environments.

​
Key capabilities

Agent mode

Delegate tasks with : the AI plans, executes with your approval and self-corrects, with built-in safety against dangerous actions.

Coder mode

Software engineering with : reads, edits, applies patches and runs tests in a loop with automatic rollback.

Quality harness

7 reliability patterns (ReAct, Reflexion, CoVe, RAG+HyDE, Self-Refine…) opt-in via /config quality.

Atomic tools

, , narrow read-only + (TodoWrite parity) + transactional. Architectural parity with Claude Code.

Smart context

Inject context with , , , , and . Your environment straight into the AI prompt.

Multiagente

14 built-in specialist agents (File, Coder, Shell, Git, Search, Planner, Reviewer, Tester…) in parallel.

Agent squad

Live run observability, shared kanban board, agent-to-agent mail, and an autonomous plan → develop → review → deliver playbook.

Customizable agents

Create personas with reusable skills in Markdown. Version-controlled with Git, shareable across teams.

K8s watcher and AIOps

Monitor K8s workloads in real time. Operator with 17 CRDs, 54+ remediation actions, log analysis, and metrics.

Durable scheduler

Cron, wait-until, DAG, and daemon mode. Jobs survive CLI restarts. Agents schedule their own follow-ups.

Conversation control

compacts history preserving what matters. and Esc+Esc go back to any point.

Hooks and web tools

Lifecycle hooks (pre/post tool, session start/end), native WebFetch and WebSearch to enrich context.

​
Current versions

ChatCLI (client)

1.223.0 — The binary you install and use in your terminal. Interactive mode, agent, coder, smart context, and remote server connection.
brew install chatcli · go install · binary

gRPC server

1.223.0 — Centralized deployment for teams. Clients connect via chatcli connect and share LLM providers, sessions, and plugins.
ghcr.io/diillson/chatcli:1.223.0

AIOps operator

1.223.0 — Autonomous incident detection, AI-powered analysis, and automated remediation for Kubernetes clusters. 17 CRDs, 54+ actions.
ghcr.io/diillson/chatcli-operator:1.223.0

Helm charts

1.223.0 — Distributed via ArtifactHub and OCI registry on GHCR.
oci://ghcr.io/diillson/charts/chatcli
oci://ghcr.io/diillson/charts/chatcli-operator