euron-coding-agent

The Python agent backend and standalone CLI behind the Euron Coding Agent VS Code extension. An open-source, model-agnostic agentic coding assistant that plans, edits code, runs commands, searches the web, reviews diffs, coordinates teams of sub-agents, and runs on cron schedules - with your approval at every step.

Provider-agnostic: works with Anthropic, OpenAI, Google Gemini, OpenRouter, Groq, Cerebras, DeepSeek, Together, Mistral, xAI, Vercel AI Gateway, Euron/Euri, Ollama, LM Studio, or any OpenAI-compatible / self-hosted endpoint.

Install

pip install euron-coding-agent
euron-agent update                            # update to the latest anytime
# optional extras:
pip install "euron-coding-agent[anthropic]"   # native Anthropic client
pip install "euron-coding-agent[mcp]"         # Model Context Protocol servers

Tip: install into a virtual environment for isolation. The client auto-adapts to each provider's API (e.g. models needing max_completion_tokens instead of max_tokens, or rejecting a custom temperature) - no model-specific config needed.

Quick start (CLI)

euron-agent                                    # interactive chat in the current folder
euron-agent run "add a /health route to app.py"
euron-agent run "summarize the repo" --json     # headless JSON output for CI/scripts
euron-agent serve --host 0.0.0.0 --port 8000    # self-host with bearer-token auth

Set a provider and key right inside the chat (no files needed):

you: /provider openai      # or gemini, anthropic, groq, ollama, ...
you: /key                  # paste your API key (stored in ~/.euron-agent)
you: create a FastAPI hello-world app with a test

Or use environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, EURI_API_KEY, ...) or a config.yaml. Built-in provider profiles mean no config file is required - just pick a provider and supply a key.

CLI commands

euron-agent                     interactive chat (remembers provider/key)
euron-agent run "<task>"        one-shot task ( --yes to auto-approve, --json headless )
euron-agent providers           list configured providers
euron-agent models              show the multi-model routing table (per-phase model + price)
euron-agent init                scaffold config.yaml + .env
euron-agent serve               run the API / WebSocket server ( --port 0 = auto )
euron-agent --team-name <name> "<task>"   multi-agent team (coordinator + specialists)
euron-agent team                list multi-agent teams
euron-agent schedule create "<name>" --cron "0 9 * * MON-FRI" --prompt "<task>"
euron-agent schedule list | run <id> | remove <id> | daemon
euron-agent plugin add <dir|zip-url> | list | remove <name>
euron-agent security            run a security audit of the codebase
euron-agent scan                fast secret + dependency vulnerability scan
euron-agent secfix              autonomous security remediation (audit→fix→verify)
euron-agent test [--all]        write + run tests (--all = whole-project suite)
euron-agent ship "<what>"       plan→build→test→scan→deploy→monitor, end to end
euron-agent deploy [--check]    deploy to a live URL (or list target readiness)
euron-agent monitor             monitoring status (errors/uptime/providers)
euron-agent heal [<url>]        self-heal: health-check, auto-rollback, error→PR
euron-agent doctor              environment self-check
euron-agent audit [--lines N]   show & verify the tamper-evident action log
euron-agent init-ci             scaffold a GitHub Actions workflow
euron-agent sessions            list saved sessions (dashboard)
euron-agent chat --session <id> | --resume   resume a past session

In-chat commands: /provider /key /model /effort /plan /execute /review /security /scan /secfix /test /testall /audit /doctor /compact /init /skills /search /usage /undo /reset /yes /help /exit, plus any custom command in .euron/commands/.

What's new (1.4.0)

• The full loop — one sentence to a live, monitored, self-healing app. euron-agent ship "build a FastAPI todo API with a Neon DB and put it live" runs plan → build → test → security-scan → deploy → monitor.
• Auto-deploy headless to Cloud Run, Cloudflare/Pages, Vercel, Netlify, Render, Fly, Railway, Docker, Kubernetes/Helm, AWS (SAM/App Runner), Azure Container Apps, plus Neon/Supabase databases — each via its CLI + a credential env var.
• Spend gate: free-tier targets deploy directly; billable ones are blocked unless you set deploy.allow_billable. Secrets are read from env, never printed, and every deploy/rollback is recorded in the audit log.
• Monitoring (euron-agent monitor): wires Sentry / OpenTelemetry / uptime and reports current errors + uptime.
• Self-healing (euron-agent heal): auto-rollback to the last known-good release on a failed health check; production errors become a fix + regression test in a PR (never auto-merged). See docs/DEPLOY.md.

What's new (1.3.0)

• Multi-model routing: assign a different model (from any provider) to each phase — planner, executor, cheap (sub-agents), verifier — via models: + routing: in config.yaml. Cost-aware strategy: auto runs cheap models where safe and auto-escalates to a stronger model when the cheaper one keeps failing, so you spend the least that still gets the job done. No provider lock-in.
• euron-agent models / /models: see which model + price runs each phase.
• Cost is now tracked per model actually used. See docs/MULTI_MODEL.md.

What's new (1.2.0)

• repo_map tool: a compact symbol/outline map of the codebase so the agent reads outlines first and full file bodies only on demand (token-saving).
• secret_scan + dependency_audit tools, and /scan / /secfix commands — find hard-coded credentials and vulnerable deps, then remediate in an autonomous audit → fix → verify loop.
• Tamper-evident audit log (.euron/audit/) — SHA-256 hash-chained record of every action; euron-agent audit / /audit verifies the chain.
• Sandbox / egress policy (sandbox:): deny-by-default command rules, allowlist, and block_network, enforced before every shell command (even in dangerous mode).
• Agent-of-agent budgets (subagent_max_calls, subagent_token_budget), model routing (router: {cheap, heavy}), an opt-in verifier sub-agent (verify_edits), and self-heal (self_heal: N).
• euron-agent doctor (environment self-check), init-ci (GitHub Actions), and Anthropic prompt caching of the static system prompt.

What's new (1.1.0)

• Drag-and-drop context: reference or drop a file, folder, or image path into chat and it is read and used automatically. Files of any reasonable length (smart head+tail truncation), folders read recursively, and images sent as multimodal blocks so vision models can actually see them. No special syntax - quoted paths, absolute/relative paths, bare filenames, and @mentions all work.
• Built-in security audit (/security, euron-agent security): vulnerability review with severity-ranked findings and concrete fixes.
• Autonomous testing: /test [target] writes and runs tests for a file/module; /testall (or euron-agent test --all) builds and runs a comprehensive test suite for the whole project and reports coverage gaps.
• Plan mode and execute mode, switchable mid-session with /plan and /execute (execute is the default).
• Earlier: auto-onboarding (.euron/ memory + skill + project doc), live status line + completion summary, dangerous (YOLO) mode, / command autocomplete, accurate per-model cost, and memory/context optimization.

Features

• Auto-onboarding (.euron/ memory + skill + project doc, created automatically).
• Agentic tool-calling loop: plan, read, search, edit, run, verify.
• 28 sandboxed tools: read/write/edit/multi_edit/create/delete files, glob, search, run_command (streaming), background processes, git status/diff/commit/ branch/push, open_pr, git worktrees, web_search, web_fetch.
• Plan mode and execute mode, sub-agents, and multi-agent teams with persistent state.
• Drag-and-drop context: read files (any length), whole folders, and images.
• Full loop: one sentence → build → test → scan → deploy to a live URL → monitor → self-heal.
• Auto-deploy to 13+ targets (Cloud Run/Cloudflare/Vercel/Netlify/Fly/Railway/Render/Docker/K8s/AWS/Azure + Neon/Supabase) with a free-tier-first spend gate.
• Multi-model routing: a different model per phase across providers, cost-aware with auto-escalation.
• Token-friendly code intelligence: repo_map outline tool, model routing, prompt caching.
• Security suite: /security, /scan, /secfix, secret_scan + dependency_audit tools.
• Tamper-evident audit log + sandbox/egress policy; sub-agent budgets, verifier, self-heal.
• Autonomous test writing/running (/test, /testall); doctor + init-ci helpers.
• Scheduled agents on cron schedules (independent of any terminal).
• MCP servers, plugins, skills, project memory (AGENTS.md), and custom commands.
• Permissions (allow/ask/deny), hooks, approvals with diffs, checkpoints, undo.
• Web search/fetch, multimodal image input, token and cost tracking, extended thinking, model fallback chains.
• Messaging notifications: Slack, Discord, Telegram, Google Chat, WhatsApp, Linear.
• Headless JSON mode and a self-hostable server with bearer-token auth.

Server

• GET /health liveness.
• GET /providers configured providers.
• POST /agent/run one-shot, non-interactive (set "auto_approve": true).
• WS /ws streaming agent with per-action approval (used by the extension).

The WebSocket init message accepts provider, model, api_key, and base_url so secrets never need to live on disk. Set EURON_AGENT_TOKEN (or serve --token) to require bearer auth.

See the project README at https://github.com/euron-tech/Euron-coding-agent for the full architecture and the VS Code extension.

License

Apache License 2.0. Copyright 2026 Euron Engage Sphere Technology Private Limited.