Crate apollo 

apollo — Lightweight agent runtime Successor to OpenClaw. Best-of-breed from ZeroClaw, NanoClaw, HiClaw.

Core traits (all swappable):

• Provider — LLM backend (Anthropic, OpenAI, Gemini, Ollama, OpenRouter, Groq)
• Channel — Communication (CLI, Telegram, Discord, Matrix, WebSocket)
• Tool — Agent capability (Shell, File I/O, Vibemania, custom)
• MemoryBackend — Persistent SurrealDB state, vector embeddings, file-based tooling

Embeddings — Vector search for semantic memory Swarms — Manager/Worker pattern for parallel execution Plugins — JSON-RPC 2.0 extensibility Cost — Token counting and billing (Phase 4) Scheduler — Cron-based task automation (Phase 4)

Re-exports

pub use agent::AgentMode;

pub use agent::AgentRunner;

pub use channels::Channel;

pub use cost::CostTracker;

pub use scheduler::Scheduler;

Modules

agent

Agent — the autonomous agent loop. Receives messages, uses tools, responds via channels. Inspired by HiClaw’s Manager/Worker pattern.

agent_http

autonomous

Autonomous coding mode.

bootstrap

Runtime/bootstrap helpers used by the CLI entrypoint.

channel_check

apollo channel-check — drive one channel against its real service.

channels

Channel abstraction — communication interfaces Enable only what you need via Cargo features.

claw_adapter

Claw adapter — migrate from OpenClaw to aclaw Maps SOUL.md, AGENTS.md config to aclaw Config Enables existing Claw workflows to run on aclaw

config

Configuration management.

context

cost

Cost tracking — token counting and billing for LLM calls Phase 4 feature: Production billing support

cron_scheduler

SurrealDB-backed Cron Scheduler — persistent scheduled tasks. Stores jobs in SurrealDB, ticks every 60s, spawns agent sessions for due jobs.

diagnostics

Shared diagnostics and security audit helpers.

embeddings

Vector embeddings for semantic memory search Uses Gemini text-embedding-004 API (free tier)

escape

Escaping for the three text formats apollo generates from paths it does not control: launchd plists, systemd units, and a generated shell script.

fs_secure

heartbeat

Heartbeat system — periodic check-ins that read HEARTBEAT.md and trigger agent actions when tasks are present.

http

mcp

MCP (Model Context Protocol) client for integrating with Codex and other MCP servers Implements JSON-RPC 2.0 over stdio transport

mcp_server

MCP server mode — exposes apollo as an MCP server over stdio or HTTP. Other AI clients can connect to prompt apollo or use its tools.

memory

Memory abstraction — persistent agent state. Inspired by ZeroClaw’s pluggable memory + NanoClaw’s per-group isolation.

plugin

Plugin system — JSON-RPC 2.0 interface + lifecycle hooks.

plugin_exec

Running tools declared by a host plugin manifest.

plugin_hosts

OpenClaw + Hermes plugin discovery (host-agnostic).

plugins

Workspace plugin manifest (Hermes/OpenClaw-style packages).

policy

Shared runtime policy controls for privileged capabilities.

process_cmd

Spawn processes from a command line without sh -c (no shell metacharacter interpretation).

prompt

System prompt builder — reads SOUL.md, USER.md, AGENTS.md, MEMORY.md, TOOLS.md, IDENTITY.md and assembles them into a system prompt for the LLM.

providers

LLM Provider abstraction — swap backends without changing agent logic. Enable only what you need via Cargo features.

redaction

scheduler

Cron scheduler — recurring task automation Phase 4 feature: Time-based task scheduling

self_update

Self-update support for repository-backed local installs.

shell_scan

Shell-aware command normalization: reduce a command to a scannable form before pattern matching.

skills

Skills system — scan SKILL.md files, match against user requests, inject matched skill instructions into the system prompt. Supports template variables and inline shell execution.

streaming_parser

Streaming tool-call parser — state-machine extraction from partial LLM output.

telegram_runtime

Telegram chat runtime extracted from the CLI entrypoint.

text

Small text utilities shared across tools and providers.

tools

Tool abstraction — agent capabilities matching OpenClaw’s tool set.

trajectory

Trajectory export for RL training data generation.

workspace_init

Workspace identity kit — minimal templates on first run.