--- name: agent-builder description: Build and configure Indemn AI agents — create agents, set prompts, add functions/tools, manage knowledge bases, and test via chat. Use when creating, configuring, or testing agents. --- # Agent Builder You build and configure AI agents using the `indemn` CLI. This skill covers the full lifecycle: create an agent, set its prompt, attach functions and knowledge bases, and test via chat. **Prerequisite:** The `indemn` CLI must be installed and authenticated. Run `indemn whoami` to verify. --- ## Workflow Create agent -> Set prompt & model -> Add functions -> Add knowledge bases -> Link KBs -> Test via chat -> Iterate. --- ## Step 1: Create the Agent Before creating, determine: name/purpose, channels (WEB, WHATSAPP, etc.), persona, functions needed, knowledge sources, and model (`indemn models [--json]`). ```bash indemn agents create --name "Acme Insurance Assistant" --channels "WEB,WHATSAPP" ``` - `--channels` defaults to `"WEB"` if omitted. Comma-separated for multiple. - Save the returned agent ID -- every subsequent command needs it. ### Other Agent Operations ```bash indemn agents list [--json] [--limit 20] [--page 2] indemn agents get [--json] indemn agents update --name "New Name" --description "Updated description" indemn agents update --webhook-url "https://example.com/webhook" indemn agents clone indemn agents delete # requires confirmation indemn agents card --output ./agent-card.pdf # branded PDF ``` --- ## Step 2: Set the System Prompt and Model ### Inline prompt ```bash indemn config set --prompt "You are an insurance assistant for Acme Corp..." ``` ### Prompt from file (preferred for long prompts) ```bash indemn config set --prompt-file ./prompts/acme-assistant.txt ``` If both `--prompt` and `--prompt-file` are provided, `--prompt-file` takes precedence. ### Set the model This controls which LLM the agent uses to generate responses — in conversations and during evaluations. The provider is auto-detected from the model name (claude -> anthropic, gemini -> google, default -> openai). This is NOT the evaluation judge model — that is set per-run via `eval run --eval-model`. ```bash indemn config set --model "claude-sonnet-4-20250514" ``` Combine prompt and model in one call: ```bash indemn config set --prompt-file ./prompts/acme.txt --model "claude-sonnet-4-20250514" ``` ### Read current configuration ```bash indemn config get [--json] indemn config get --field prompt indemn config get --field model indemn config get --field greeting ``` --- ## Step 3: Add Functions (Tools) Functions are tools the agent can invoke during conversation (e.g., generate a quote, transfer to a human, look up policy details). ### Create a function ```bash indemn functions create --type ACTION --name "get_quote" --description "Generate an insurance quote" ``` ### Add parameters to a function ```bash indemn functions params add \ --name "state" --type string --description "US state abbreviation" --required indemn functions params add \ --name "coverage_amount" --type number --description "Coverage amount in dollars" indemn functions params add \ --name "has_claims" --type boolean --description "Whether applicant has prior claims" ``` Parameter types: `string`, `number`, `boolean`. Use `--required` for mandatory params. ### Manage parameters ```bash indemn functions params list [--json] indemn functions params update --description "Updated desc" --required indemn functions params delete ``` ### Import functions from an OpenAPI spec ```bash indemn functions import --url "https://api.example.com/openapi.json" ``` This parses the spec and creates functions with parameters automatically. ### Other function operations ```bash indemn functions list [--json] [--limit 10] [--page 1] indemn functions update --name "updated_name" --description "New desc" indemn functions test [--json] indemn functions export [--json] indemn functions delete # requires confirmation indemn functions master-list [--json] # available function templates ``` --- ## Step 4: Create and Link Knowledge Bases Knowledge bases provide the agent with reference information. Three data source types: QNA pairs, URLs, and files. ### Create a knowledge base ```bash indemn kb create --name "Acme FAQ" --type QNA indemn kb create --name "Product Pages" --type URL indemn kb create --name "Policy Documents" --type FILE ``` Type defaults to `QNA` if omitted. Must be uppercased: `QNA`, `URL`, or `FILE`. ### Add data -- QNA pairs ```bash indemn kb data add --question "What is the deductible?" --answer "The standard deductible is $500." ``` ### Add data -- URL scraping ```bash indemn kb data add --source-url "https://acme.com/faq" ``` Scrapes the URL and ingests the content. ### Add data -- File upload ```bash indemn kb data add --file ./documents/policy-handbook.pdf ``` Uploads the file and polls until processing is complete. ### Bulk import QNA pairs from a file ```bash indemn kb import --file ./qna-pairs.json ``` The file must be a JSON array of `{question, answer}` objects: ```json [ {"question": "What is a deductible?", "answer": "The amount you pay before coverage kicks in."}, {"question": "What is a premium?", "answer": "The amount you pay for your insurance policy."} ] ``` Each pair is added individually. Progress is shown during import. ### Link knowledge bases to an agent ```bash indemn kb link indemn kb link # link multiple at once ``` The link command is variadic -- pass multiple KB IDs to link them all in one call. ### Unlink a knowledge base ```bash indemn kb unlink ``` ### Other KB operations ```bash indemn kb list [--json] [--limit 20] [--page 1] indemn kb get [--json] indemn kb update --name "Updated Name" indemn kb delete # requires confirmation indemn kb data list [--json] [--limit 20] [--page 1] # shows Total, Page X of Y indemn kb data update --question "New Q" --answer "New A" indemn kb data update --source-url "https://new-url.com" indemn kb data update --content "Replacement content" indemn kb data delete # requires confirmation indemn kb export --format csv # or xlsx indemn kb import --file ./qna-pairs.json # bulk import QNA pairs ``` --- ## Step 5: Test via Chat ### Interactive mode ```bash indemn chat ``` Opens a live chat session. Type messages, see responses in real time. Ctrl+C to exit. ### Single message (scripting / quick check) ```bash indemn chat --message "What types of insurance do you offer?" ``` Sends one message, prints the response, exits. ### Resume a previous session ```bash indemn chat --session ``` Continues an existing conversation thread with its history intact. --- ## Step 6: Iterate After testing, refine: update the prompt (`config set --prompt-file`), add functions (`functions create`), add/update KB data (`kb data add` / `kb data update`), or switch models (`config set --model`). Repeat steps 2-5 until the agent performs correctly. ## Complete Build Example ```bash # 1. Create agent indemn agents create --name "Acme Wedding Insurance" --channels "WEB" # -> agent_id: abc123 # 2. Set prompt and model indemn config set abc123 --prompt-file ./prompts/wedding.txt --model "claude-sonnet-4-20250514" # 3. Add functions indemn functions create abc123 --type ACTION --name "get_quote" --description "Generate a wedding insurance quote" # -> func_id: func456 indemn functions params add abc123 func456 --name "event_date" --type string --description "Wedding date" --required indemn functions params add abc123 func456 --name "guest_count" --type number --description "Number of guests" --required indemn functions params add abc123 func456 --name "venue_cost" --type number --description "Total venue cost in dollars" # 4. Create and populate knowledge base indemn kb create --name "Wedding FAQ" --type QNA # -> kb_id: kb789 indemn kb data add kb789 --question "What does wedding insurance cover?" --answer "Coverage includes venue damage, vendor no-shows, severe weather, and illness." indemn kb data add kb789 --source-url "https://acme.com/wedding-insurance/faq" # 5. Link KB to agent indemn kb link abc123 kb789 # 6. Test indemn chat abc123 --message "Hi, I'm getting married in June and want to protect my investment." indemn chat abc123 # interactive session for deeper testing ``` --- ## CLI Quick Reference | Group | Command | Purpose | |-------|---------|---------| | **Agents** | `agents list / get / create / update / delete` | CRUD operations | | | `agents clone ` | Duplicate an agent | | | `agents card [--output ]` | Generate branded PDF | | **Config** | `config get [--field prompt\|model\|greeting]` | Read configuration | | | `config set [--prompt\|--prompt-file\|--model]` | Write configuration | | **Functions** | `functions list / create / update / delete` | CRUD operations | | | `functions test ` | Test a function | | | `functions export ` | Export all functions | | | `functions import --url ""` | Import from OpenAPI | | | `functions master-list` | Available templates | | | `functions params list / add / update / delete` | Manage parameters | | **KB** | `kb list / get / create / update / delete` | CRUD operations | | | `kb data list / add / update / delete` | Manage KB entries | | | `kb link ` | Link KB(s) to agent | | | `kb unlink ` | Unlink KB from agent | | | `kb export [--format csv\|xlsx]` | Export KB data | | | `kb import --file ` | Import from JSON | | **Chat** | `chat ` | Interactive session | | | `chat --message ""` | Single message mode | | | `chat --session ` | Resume session | | **Other** | `whoami` | Show authenticated user | | | `models` | List available LLMs | All commands accept `--json` for machine-readable output. List commands accept `--limit ` and `--page `. Delete commands require confirmation.