Daneel

"The Laws of Robotics are not suggestions." — R. Daneel Olivaw

Daneel is a Go library for building production-grade AI agents and multi-agent systems.
Zero external dependencies — just the Go standard library.

---

Features

Category	Capabilities
Core	Typed tool definitions, permission policies, handoff between agents, session management
Providers	OpenAI, Anthropic, Google Gemini, Ollama — with streaming (ChatStream) and circuit breaker
Memory	Sliding-window, summarisation, vector store, file-backed persistence
Context	Automatic context window management — sliding trim, summarize, error strategies (WithMaxContextTokens)
Multi-agent	Chain, parallel, router, orchestrator, FSM workflows — max handoff depth protection
Platforms	Slack, Telegram, Twitter/X, WhatsApp, GitHub, Email
Protocols	Model Context Protocol (MCP) client & server, WebSocket connector
Observability	OpenTelemetry tracing & metrics, bridge.BridgeMetrics interface, fine-tune dataset collection
Local AI	WithLocalStack auto-detects Ollama, ollama.NewEmbedder for local embeddings, zero-config offline mode
Docker	Multi-stage distroless Dockerfile, docker-compose.yml (Ollama + Daneel), daneel doctor health check
Operations	Multi-tenant quotas (scoped sessions), billing / cost tracking, budget alerts, cron scheduling
Resiliencia	Circuit breaker, fail-fast tool strategy (WithFailFast), composable run hooks
Config	JSON config loading (LoadConfig), ${ENV} expansion, BuildAgents / BuildPlatforms
Validation	RunStructured[T] with JSON Schema validation (required, enum) and auto-retry
Human-in-loop	Tool-call approval, A/B testing, LLM-as-judge evaluation
CLI	daneel binary — run agents, listen on platforms, trigger fine-tuning

---

Installation

go get github.com/Rafiki81/daneel

Requires Go 1.24+.

---

Quick Start

package main

import (
    "context"
    "fmt"
    "log"
    "strings"

    "github.com/Rafiki81/daneel"
)

type WeatherParams struct {
    City string `json:"city" desc:"City name"`
}

func main() {
    // 1. Define a typed tool
    weatherTool := daneel.NewTool("get_weather", "Get current weather for a city",
        func(ctx context.Context, p WeatherParams) (string, error) {
            temps := map[string]string{"madrid": "28°C ☀️", "tokyo": "22°C ⛅"}
            city := strings.ToLower(p.City)
            if t, ok := temps[city]; ok {
                return t, nil
            }
            return "20°C 🌡️", nil
        },
    )

    // 2. Create an agent
    agent := daneel.New("assistant",
        daneel.WithInstructions("You are a helpful weather assistant."),
        daneel.WithModel("gpt-4o"),
        daneel.WithTools(weatherTool),
        daneel.WithMaxTurns(5),
    )

    // 3. Run it
    result, err := daneel.Run(context.Background(), agent, "Weather in Madrid?")
    if err != nil {
        log.Fatal(err)
    }
    fmt.Println(result.Output)
}

---

Examples

Example	Description
examples/quickstart	Single agent with a tool and interactive REPL
examples/multi-platform	Same agent running on Slack + Telegram simultaneously
examples/github-reviewer	PR review bot using the GitHub platform
examples/slack-assistant	Slack bot with memory and file upload support
examples/twitter-bot	Automated Twitter/X agent
examples/permissions	Fine-grained tool permission policies

---

Local-First AI

Daneel auto-detects a local Ollama instance and falls back to it when no OPENAI_API_KEY is set:

// Explicit local stack — uses Ollama for chat + embeddings
agent := daneel.New("assistant",
    daneel.WithLocalStack("llama3.2", "nomic-embed-text"),
)

// Auto-detect (no OPENAI_API_KEY → tries Ollama at localhost:11434)
result, err := daneel.Run(ctx, agent, "Hello")

For embeddings only:

import "github.com/Rafiki81/daneel/provider/ollama"

e := ollama.NewEmbedder(ollama.EmbedModel("nomic-embed-text"))
vec, err := e.Embed(ctx, "semantic search query")

Check your local stack health with:

daneel doctor

---

Docker & Deployment

A production-ready multi-stage image is included:

# Build the image
docker build -t daneel .

# Run Daneel + Ollama together
docker compose up

The Dockerfile uses golang:1.24-alpine builder → gcr.io/distroless/static-debian12:nonroot for a minimal, non-root runtime image.

---

Package Index

Core

Package	Description
github.com/Rafiki81/daneel	Agent, Tool, Run, RunResult, Registry, Connector
github.com/Rafiki81/daneel/content	Multi-modal content types (text, image, file)
github.com/Rafiki81/daneel/approval	Human-in-the-loop approval for tool calls
github.com/Rafiki81/daneel/bridge	Point-to-point agent bridging

Providers

Package	Description
github.com/Rafiki81/daneel/provider	Circuit breaker (CircuitBreaker), fallback, round-robin provider wrappers
github.com/Rafiki81/daneel/provider/openai	OpenAI (GPT-4o, o1, …) — streaming support
github.com/Rafiki81/daneel/provider/anthropic	Anthropic (Claude 3.x, …) — streaming support
github.com/Rafiki81/daneel/provider/google	Google Gemini — streaming support
github.com/Rafiki81/daneel/provider/ollama	Ollama (local models) — streaming support

Memory

Package	Description
github.com/Rafiki81/daneel/memory	Sliding-window + summary memory
github.com/Rafiki81/daneel/memory/store	In-memory vector store
github.com/Rafiki81/daneel/session	Persistent session store (memory, file) with automatic cleanup

Workflows

Package	Description
github.com/Rafiki81/daneel/workflow	Chain, parallel, router, orchestrator, FSM

Platforms & Connectors

Package	Description
github.com/Rafiki81/daneel/platform/slack	Slack client & tools
github.com/Rafiki81/daneel/platform/telegram	Telegram client & tools
github.com/Rafiki81/daneel/platform/twitter	Twitter/X client & tools
github.com/Rafiki81/daneel/platform/whatsapp	WhatsApp client & tools
github.com/Rafiki81/daneel/platform/github	GitHub client & tools
github.com/Rafiki81/daneel/platform/email	Email client & tools
github.com/Rafiki81/daneel/connector/*	High-level connectors for each platform
github.com/Rafiki81/daneel/ws	WebSocket server and connector (stdlib only)

Protocols

Package	Description
github.com/Rafiki81/daneel/mcp	Model Context Protocol client & server

Knowledge & Scheduling

Package	Description
github.com/Rafiki81/daneel/knowledge	Document ingestion, chunking, RAG retrieval
github.com/Rafiki81/daneel/cron	Cron-style scheduled agent runs

Experimentation

Package	Description
github.com/Rafiki81/daneel/experiment	A/B testing, LLM-as-judge, metrics
github.com/Rafiki81/daneel/finetune	Fine-tune dataset collection & evaluation

Operations

Package	Description
github.com/Rafiki81/daneel/tenant	Multi-tenant quota management & scoped sessions
github.com/Rafiki81/daneel/billing	Cost tracking, pricing tables, budget alerts
github.com/Rafiki81/daneel/pubsub	Publish/subscribe message bus with agent tools
github.com/Rafiki81/daneel/trace	OpenTelemetry tracing and metrics

CLI

Package	Description
github.com/Rafiki81/daneel/cmd/daneel	CLI — agents, tools, run, listen, finetune, doctor

---

CLI

Install the daneel binary:

go install github.com/Rafiki81/daneel/cmd/daneel@latest

daneel agents list              # list registered agents
daneel agents describe <name>   # show agent details
daneel tools list               # list registered tools
daneel run <agent> "prompt"     # run an agent once
daneel run <agent>              # interactive REPL
daneel listen --slack           # start Slack listener
daneel finetune --dataset data.jsonl --base gpt-4o
daneel doctor                   # check Ollama, API keys, Go toolchain
daneel doctor --json            # machine-readable output

---

Multi-Agent Workflow Example

import "github.com/Rafiki81/daneel/workflow"

result, err := workflow.Chain(ctx, input,
    researchAgent,
    writerAgent,
    editorAgent,
)

---

Contributing

See CONTRIBUTING.md. All contributions welcome.

---

License

MIT © 2026 The Daneel Authors