Raxol

Recursively, axol. Forever FOSS.

Write one app. Render it to a terminal, a browser, an SSH session, or an agent.

Your application is a single TEA module (init, update, view) running as an OTP GenServer. Raxol renders that module to four surfaces from one codebase:

                          +---> Terminal (termbox2 NIF)
                          |
  TEA module (GenServer) -+---> Browser (Phoenix LiveView)
                          |
                          +---> SSH (Erlang :ssh)
                          |
                          +---> Agent (MCP tools)

The interesting part is the runtime. Your app gets crash isolation per Component, hot code reload without restart, distributed clustering with CRDTs, and an agent surface where LLMs interact with structured Component trees instead of scraping pixels. Those are BEAM properties, from a VM built for systems that can't go down, can't lose state, and hot-swap code while running.

Bubble Tea, Ratatui, and Textual are excellent renderers. A2UI and AG-UI define agent-UI wire formats. Raxol is the runtime that renders all four surfaces from one source module. See Why OTP for the framework comparison, and Why Raxol for how the runtime compares to Python agent stacks like Hermes and Omnigent.

Agents

Raxol is a runtime for agents as much as for humans. Every interactive Component automatically exposes MCP tools (Button gives click, TextInput gives type_into/clear/get_value), and a focus lens filters to roughly 15 relevant tools per interaction. Where A2UI and AG-UI define how agents talk to UIs at the wire level, raxol generates the UI and the agent surface from one Component tree: same source, two projections.

import Raxol.MCP.Test
import Raxol.MCP.Test.Assertions

session = start_session(MyApp)

session
|> type_into("search", "elixir")
|> click("submit")
|> assert_component("results", fn c -> c[:content] != nil end)
|> stop_session()

mix mcp.server starts the MCP server on stdio for Claude Code integration, and mix raxol.code is an interactive terminal coding agent (the axol face) with every mutating tool call gated by an ALLOW/ASK/DENY authorization engine. See the Coding Agent.

Code is the coding-agent product, in two hands-on surfaces: mix raxol.code is the interactive terminal TUI (the axol face ≡··≡), and mix raxol.p is its headless twin (prompt in on argv, answer to stdout, contract events to stderr) for pipes and CI. Every mutating tool call is gated by an ALLOW/ASK/DENY authorization engine. From a clone, one setup command and one launch:

(cd packages/raxol_agent && mix deps.get)   # once
bin/raxol-code                              # the TUI, your cwd as the workspace

No API key configured? The TUI opens on a provider wizard instead of failing. /inspect (or mix raxol.inspect from packages/raxol_agent) shows every config source the agent will use in the current directory.

Sessions are durable. Each session journals its events to disk, so --continue and --resume <id> restore the model context and the scrollback together, --replay <id> prints a transcript straight from the journal (--to-offset N stops at an offset), /rewind drops back to an earlier turn, and /share mints a signed 24-hour link to a read-only transcript that follows the session live (needs RAXOL_SHARE_SECRET and a host mounting Raxol.Agent.Code.ShareLive).

The same TUI serves over SSH: mix raxol.code --ssh --ssh-tenants /srv/tenants hosts many users from one daemon, each behind their own public key with their own cwd jail, session store, and spending identity (ssh <you>@your-host -p 2222 is the whole client). Single-tenant (--authorized-keys) and hosted deployment (RAXOL_SSH_CODE=true) are in Coding Agent.

Two unattended surfaces run the same agent. bin/raxol-acp (or mix raxol.acp) serves it over the Agent Client Protocol on stdio, for editors that spawn an agent themselves; Raxol.Agent.Harness.McpTools registers harness_start_session, harness_send_prompt, harness_read_transcript, and harness_list_sessions with the MCP registry, so a session started by an MCP client resumes later in the TUI. The ACP surface runs the full toolset with every sensitive call gated on a session/request_permission round trip, fail-closed on the decision: a client that refuses, times out, or does not implement permissions denies the write and keeps reading. The MCP surface stays read-only: write_file, edit_file, and bash are absent where nobody is there to answer an approval prompt.

Every one of those surfaces sits on the Harness, the agent-session engine: a durable event journal (Raxol.Agent.Journal), a typed event/command contract (Raxol.Agent.Contract), and surface state that is a pure fold over the event stream (Raxol.Harness.Projection), with staged interrupt, steer, and spend/blast-radius gates underneath. The same engine can supervise external agent CLIs (claude, cursor) as readily as Raxol's own loop. See Harness architecture.

The agent subsystems ship as standalone packages:

• Pay (raxol_payments): wallets, ledger-enforced spending limits, and transparent auto-pay when an agent hits an HTTP 402, across five protocols (x402, MPP, Xochi cross-chain, Permit2, Riddler).
• Earn (raxol_earn): the sell side. Declare an offering, implement two callbacks, and a buyer agent discovers it, escrows funds, and settles on-chain through the Virtuals ACP job lifecycle (request, negotiation, transaction, evaluation, completed), one supervised process per job. Pre-alpha.
• Improve (raxol_agent): a solved task becomes a reusable SKILL.md. A background reviewer runs on a cheap model after each turn, writing durable memory and new skills without spending the live turn's latency or context.
• Reach (raxol_gateway): one adapter contract to many chat platforms, process-per-chat sessions, DM pairing for authorization, and /handoff to move a conversation across platforms with its history intact.
• Orchestrate (raxol_symphony): an OTP port of OpenAI Symphony that polls a tracker, isolates each issue in its own workspace, and runs a coding agent, feeding six surfaces (terminal, LiveView, MCP, Telegram, Watch, JSON API) from one snapshot.
• Bridge (raxol_agent_client_protocol): Elixir/OTP implementation of the Agent Client Protocol: the JSON-RPC 2.0 wire protocol between code editors and AI coding agents (the protocol Zed and a growing ecosystem speak). Bidirectional agent/client roles, pluggable transports (stdio, in-process), and durable resumable sessions (offset-based reattach/replay) as a vendor extension. Zero raxol-internal deps. Pre-alpha.

Install

# mix.exs
def deps do
  [{:raxol, "~> 2.7"}]
end

Or generate a new project:

mix raxol.new my_app

With Nix, nix develop drops you into a shell with the full BEAM and NIF toolchain (no local Elixir install required):

nix develop github:DROOdotFOO/raxol   # dev shell with elixir, erlang, NIF + speech deps

Try it

Nothing to install: raxol.io runs the same catalog in a browser, one LiveView per demo. That surface is this repo's web/ app; see web/README.md to run it yourself. In a terminal:

git clone https://github.com/DROOdotFOO/raxol.git
cd raxol && mix deps.get
mix raxol.playground          # 42 live demos, browse/search/filter

The flagship demo is a live BEAM dashboard with scheduler utilization, memory sparklines, and a process table:

mix run examples/demo.exs

See examples/README.md for the full learning path, including agent examples, swarm demos, and the REPL.

Headless environment (CI, containers, agents)? The whole build-and-test path needs no tty:

mix local.hex --force        # fresh machines and CI: install Hex without a prompt
mix deps.get
mix compile                  # termbox2 NIF needs make + a C compiler
SKIP_TERMBOX2_TESTS=true MIX_ENV=test mix test --exclude slow --exclude integration --exclude docker
MIX_ENV=test mix raxol.rate  # RATE: render-determinism golden suite

Prerequisites, the quality gate, and constrained-sandbox notes are in Development.

Performance

Full frame in 5.0ms on Apple M1 (Elixir 1.20 / OTP 29), 31% of the 60fps budget.