openagentd 0.6.0

Self-hosted AI agents — streaming chat, tool use, persistent memory, multi-agent teams

OpenAgentd

Your on-machine multi-agent system. A long-running local service with a web cockpit, persistent memory, and a team of agents that coordinate to get real work done. Everything stays on your hardware.

Documentation

---

What you get

A cockpit, not a chat box. Command palette (Ctrl+P), drag-and-drop files, full-screen image viewer, and an inspector that shows every tool call and what came back.

Agents that can actually do things. Read and write files, run shell commands, search the web, generate images and video, manage todos, schedule tasks. Add more via a skill .md or any MCP server.

A workspace the agent shares with you. Every file the agent touches shows up in a side panel — browse, preview, download. In /coding, open a server-local project folder and review the live file tree plus current git diff from the same drawer.

Persistent memory you can edit. Three-tier wiki: session notes, synthesised topics, and a USER.md injected into every prompt. Browse and edit it from the Wiki panel.

Run a team, not just one agent. A lead agent spawns specialist instances on demand (executor#1, executor#2, ...), coordinates through an async mailbox, and can grant/revoke member tools, skills, or MCP servers at runtime. Watch live agents in the default split view — or switch to a single unified view.

Use it as a coding cockpit. Coding mode ships with a workspace-aware team (coding/openagentd, coding/executor, coding/explorer, coding/consultant) that can inspect a local codebase, make changes, run checks, and keep files/diffs visible while it works.

Voice input, transcribed locally. Click the mic button to record, click again to stop. The recording is transcribed on-device via Whisper and inserted into the chat input for review — nothing leaves your machine. Configure in speech.yaml or enable via Settings → Voice.

Schedule it and walk away. Cron, interval, or one-shot schedules. Results appear when you come back.

Track work as a board. todo_manage gives the lead and members a shared task board with assignment, claims, dependencies, priorities, and live UI updates.

See exactly what the agent is doing. Built-in OTel dashboard — token usage, latency, trace waterfall. No third-party SaaS, all local.

Pick your model, no lock-in. 14 providers — Gemini, OpenAI, OpenRouter, Bedrock, Grok, DeepSeek, Ollama, and more. Switch with one line in your agent config.

---

Why OpenAgentd

	openagentd	opencode	openclaw	hermes-agent
UI	Web cockpit	Terminal	Messaging apps	Messaging / CLI
Memory	3-tier wiki, cross-session, editable	Session only	Session only	Cross-session (FTS5)
Image / video	Multi-provider images + native video	—	Via plugins	Images, no video
Hot-reload	Everything, no restart	Restart required	Partial	MCP only
Self-modification	Agent edits its own config	—	Partial	Persona + skills
Telemetry	Built-in OTel dashboard	—	—	—
Embed / API	First-class REST + SSE	Protocol only	Channel-shaped	Channel-shaped

Full breakdown: documents/docs/comparison.md.

---

Quick start

# macOS / Linux
uv tool install openagentd        # recommended
brew tap lthoangg/tap && brew install openagentd
curl -fsSL https://raw.githubusercontent.com/lthoangg/openagentd/main/install.sh | sh

# Windows
irm https://raw.githubusercontent.com/lthoangg/openagentd/main/install.ps1 | iex

# Docker
git clone https://github.com/lthoangg/openagentd.git
cd openagentd && cp .env.example .env && docker compose up -d

openagentd init   # pick provider + API key, install default agents
openagentd        # http://localhost:4082

Other install options (pip, pipx, from source) — see documents/docs/install.md.

---

Migrate from OpenClaw or Hermes Agent

Import existing identity and context Markdown files into one OpenAgentd lead agent:

openagentd migrate openclaw --from ~/.openclaw/workspace --model openai:gpt-5.5
openagentd migrate hermes --from ~/.hermes --model openai:gpt-5.5

Existing agent files are not overwritten unless --force is passed. See documents/docs/configuration.md for supported source files.

---

Providers

Switch models with a single line in your agent's .md config file. Every provider uses the provider:model format.

Provider	Format	Auth
Google Gemini	googlegenai:gemini-3.1-flash	GOOGLE_API_KEY
Google Vertex AI	vertexai:gemini-3-flash-preview	VERTEXAI_API_KEY or GCP creds
OpenAI	openai:gpt-5.5	OPENAI_API_KEY
OpenRouter	openrouter:qwen/qwen3.6-plus:free	OPENROUTER_API_KEY
ZAI / GLM	zai:glm-5-turbo	ZAI_API_KEY
xAI Grok	xai:grok-4.20	XAI_API_KEY
DeepSeek	deepseek:deepseek-v4-flash	DEEPSEEK_API_KEY
AWS Bedrock	bedrock:anthropic.claude-sonnet-4-6	AWS profile / default chain
NVIDIA NIM	nvidia:stepfun-ai/step-3.5-flash	NVIDIA_API_KEY
GitHub Copilot	copilot:gpt-5.4-mini	openagentd auth copilot
OpenAI Codex	codex:gpt-5.5	openagentd auth codex
9Router (local)	router9:cc/claude-sonnet-4-5	ROUTER9_API_KEY (optional ROUTER9_BASE_URL)
CLIProxyAPI (local)	cliproxy:gemini-2.5-pro	CLIPROXY_API_KEY (optional CLIPROXY_BASE_URL)
Ollama (local + cloud)	ollama:llama3.2 · ollama:kimi-k2.6-cloud	none (cloud: ollama signin)

Set a fallback_model in your agent config for automatic failover on rate limits or 5xx errors.

---

Built-in tools

Category	Tools
Filesystem	read, write, edit, ls, glob, grep, rm
Shell	shell, bg (background processes)
Web	web_search, web_fetch
Memory	wiki_search, note
Generation	generate_image, generate_video
Scheduling	schedule_task
Tasks	todo_manage
Team coordination	team_message, team_manage, team_configure (teams only)
Utility	date, skill

Add any MCP server to expose more tools without writing code.

---

Agents and teams

OpenAgentd ships with one lead agent and three member blueprints:

Agent	Role	Specialty
openagentd	Lead	Coordinates the team, receives user messages, spawns members, delegates
consultant	Member blueprint	Architecture reviews, debugging, design decisions (high thinking)
executor	Member blueprint	File creation, builds, shell commands, tangible artifacts
explorer	Member blueprint	Web research, codebase exploration, information gathering

Configure your team by editing .md files in your config directory. Exactly one agent must have role: lead; the rest are member blueprints. The lead uses team_manage to spawn/dismiss live instances (executor#1, explorer#1), team_message to delegate and collect results, and team_configure to grant or revoke a member's skills, tools, or MCP servers without restarting.

Fresh installs also seed a separate coding team under agents/coding/. Open /coding to select a server-local project folder and start workspace-aware sessions; Settings shows those agents as coding/openagentd, coding/executor, coding/explorer, and coding/consultant.

Agent config at a glance

---
name: my-agent
role: member
description: Handles deep research tasks
model: googlegenai:gemini-3.1-flash
thinking_level: high
fallback_model: openrouter:qwen/qwen3.6-plus:free
tools:
  - web_search
  - web_fetch
  - read
  - note
skills:
  - web-research
mcp:
  - context7
summarization:
  token_threshold: 80000
  keep_last_assistants: 2
---

System prompt goes here.

Member instances are created lazily from role: member configs. Re-spawning an explicit handle restores its history for the current lead session; dismissing an instance only removes it from the live roster.

---

Memory

Three tiers, all editable:

1. USER.md — Always injected into every system prompt. Edit it directly to give the agent standing context about you, your projects, or your preferences.
2. Topics — Synthesised knowledge base, BM25-searchable via wiki_search.
3. Session notes — Per-session notes the agent appends to via the note tool.

The dream agent runs on a cron schedule, reads unprocessed session notes, synthesises new topic files, and updates the wiki index — turning ephemeral conversation into durable memory without any action on your part.

---

Voice input

Click the mic button in the chat input to record. Click again to stop. The recording is transcribed on-device using Whisper and inserted into the input for review — you still press Send manually. Nothing leaves your machine.

Enable it:

1. Open Settings → Voice and toggle it on (or edit ~/.config/openagentd/speech.yaml directly).
2. Install the local transcription extra once:

uv sync --extra voice-local
# or, for tool installs:
uv tool install "openagentd[voice-local]"

speech.yaml reference:

voice:
  enabled: true
  model: local:base    # local:base / local:small / local:medium
  language: auto       # or a BCP-47 code: "en", "fr", "ja", …
  max_file_mb: 25

The file is hot-reloaded on change — no server restart needed. V1 is local-only (local:*). No TTS, no auto-send, no silence auto-stop.

---

Scheduler

Create tasks that run on a schedule or fire once at a specific time:

• Cron — standard five-field cron expressions
• Interval — every N seconds, minutes, or hours
• At — one-shot at an exact datetime

Tasks appear in the /scheduler panel. Pause, resume, or trigger them manually from the UI or via the REST API.

---

Observability

OpenAgentd exports OpenTelemetry spans to local JSONL partitions and serves a built-in dashboard at /telemetry:

• Summary — token usage, error rates, latency distribution, model breakdown
• Trace explorer — full span waterfall per session, filterable by date range
• Prometheus endpoint — /metrics for external scraping

No external collector required. All data stays on your machine.

---

Skills

Skills are .md files that inject domain-specific instructions into an agent's context on demand. They ship separately from agent configs, so one skill can be reused by any agent.

Included skills:

Skill	Purpose
self-healing	Agent edits its own config (model, tools, skills, summarization thresholds)
mcp-installer	Install new MCP servers from the UI or by description
skill-installer	Install new skills from a URL or from scratch
plugin-installer	Install agent plugins
web-research	Structured web research methodology with source citation

Add your own by dropping a SKILL.md file into {config_dir}/skills/{name}/ or via the /settings/skills UI.

---

MCP servers

OpenAgentd ships with Context7 pre-configured. Add any MCP server via the /settings/mcp panel or by editing mcp.json directly. Changes are hot-reloaded without a restart.

{
  "servers": {
    "my-server": {
      "command": "npx",
      "args": ["my-mcp-package"],
      "env": { "API_KEY": "${MY_API_KEY}" }
    }
  }
}

---

Sandbox and permissions

Filesystem sandbox — A denylist blocks access to OpenAgentd's own data, state, and cache directories. Add your own glob patterns (**/.env, **/secrets/**) in sandbox.yaml. Changes take effect immediately, no restart needed.

Permission system — By default, tools auto-approve and log. Switch to interactive mode to block on sensitive operations and reply per-request with once, always, or reject. Permission decisions are persisted and replayed across turns.

---

Documentation

Section	Contents
Install	pip, uv, Homebrew, Docker, source
CLI reference	Every openagentd subcommand
Configuration	Env vars, agent .md files, providers, tools, skills, sandbox
Architecture	C4 diagrams, agent loop, SSE protocol
API reference	HTTP endpoints, SSE events, file handling
Agent engine	Loop, hooks, tools, teams, context, summarization
Lazy team members	Spawn/dismiss member instances, blueprint#N handles, history restore
Workspace Files	Session files, coding Files & Diff, previews, downloads
Todos popover	Task board, assignments, claims, dependencies, live updates
Tool rendering	Tool call/result UI and custom renderers
Comparison	How OpenAgentd compares to opencode, openclaw, hermes-agent
Troubleshooting	Common install and runtime issues
Guidelines	Code style, testing patterns, workflow (contributors)

---

Star History

---

Contributing

See CONTRIBUTING.md for setup, workflow, and PR guidelines.

Security

See SECURITY.md for the trust model and how to report vulnerabilities.

License

Apache License 2.0. Free for personal, research, and commercial use.