FAQ

What LLM providers does Forge support?

Forge supports OpenAI, Anthropic, Google Gemini, Ollama (local models), and custom OpenAI-compatible endpoints. You can configure fallback chains for automatic failover between providers.

Can I use Forge without an API key?

Yes. Use Ollama with a locally-hosted model:

model:
  provider: ollama
  name: llama3

No API key needed — Ollama runs models entirely on your machine.

How do I add a skill?

# Browse available skills
forge skills list

# Add a skill from the registry
forge skills add tavily-research

# Validate requirements
forge skills validate

See Embedded Skills for the full catalog.

What's the difference between skills and tools?

Skills are high-level capability descriptions defined in SKILL.md files. They describe what an agent can do and compile into container artifacts during forge build.

Tools are individual functions the LLM can call at runtime — like web_search, http_request, or cli_execute. Skills can register their own tools (script-backed) or use existing builtin tools (binary-backed).

How does Forge secure my data?

Forge provides multiple security layers:

• Egress controls — Restrict which domains agents can access (allowlist, deny-all, or dev-open modes)
• Encrypted secrets — AES-256-GCM encryption with Argon2id key derivation
• Execution sandboxing — Binary allowlists, argument validation, environment isolation
• Content guardrails — PII detection, jailbreak protection, secret scanning
• Build signing — Ed25519 signatures for supply chain integrity
• Audit logging — Structured NDJSON events for every security-relevant action

See Security Overview for the full architecture.

Can I run Forge in air-gapped environments?

Yes. Use Ollama as the LLM provider and set egress mode to deny-all:

model:
  provider: ollama
  name: llama3
egress:
  mode: deny-all

Pre-install all binary dependencies in the container image. No outbound network access is required.

How do I connect Slack or Telegram?

# Add a channel adapter
forge channel add slack
forge channel add telegram

# Run with channels
forge run --with slack,telegram

Both channels use outbound-only connections — no public URLs or webhooks required. See Channels for setup instructions.

Does Forge support multi-turn conversations?

Yes. Session persistence is enabled by default — conversations are saved to disk and recovered automatically. Sessions older than 30 minutes are discarded to prevent poisoned context.

See Memory System for details.

Can I schedule recurring tasks?

Yes. Configure schedules in forge.yaml or create them dynamically at runtime:

schedules:
  - id: daily-report
    cron: "@daily"
    task: "Generate daily status report"
    channel: telegram

See Scheduling for details.