Product changelog
Track new features, improvements, and updates across the Calljmp platform. From agent runtime APIs and observability tools to CLI enhancements and REST endpoints.
March 2026
REST API: Live WebSocket Updates
Stream real-time agent execution state to connected clients with WebSocket subscriptions.
- GET /target/v1/live/{key} — connect to live execution updates
- Receive execution state changes as JSON messages
- Auto-reconnect with exponential backoff
REST API: Agent Cancellation
Cancel running or pending agent executions via REST API.
- DELETE /target/v1/agent/{runId} — cancel execution
- Updates status to canceled
- Returns HTTP 200 on success
Web Scraping API
Extract content from web pages with text, HTML, or structured field extraction.
- web.scrape(url, format) supports text and HTML formats
- CSS selector extraction with attribute and text filters
- Automatic consent banner detection and handling
- Full page scrolling with human-like variation
@calljmp/web v0.0.2 Released
WebSocket-based Agent client for embedding real-time agent connections in web applications.
- Calljmp, Agents, Agent classes for client initialization
- WebSocket auto-reconnect with up to 10 retry attempts
- Typed message protocol for send/receive
- Development mode with custom endpoint support
Portals: Agent Web Apps
Share agents as chat-based web applications with custom subdomains and access controls.
- Create portals from dashboard — link multiple agents
- Custom domain support (agent-name.portal.calljmp.com)
- Chat interface for end users to interact with agents
- Access control and form input configuration
February 2026
Slack Integration
Post messages and rich content to Slack channels from agents.
- integrations.slack.postMessage(channel, text, blocks)
- Support for Slack Block Kit rich formatting
- Setup via dashboard OAuth flow
- Automated channel posting without manual intervention
Workflow Suspension & Human-in-the-Loop
Pause agent execution to request human approval or input before continuing.
- workflow.suspend(options) — pause execution with timeout
- Wait for external resumption before proceeding
- Optional custom reason message
- Resumable via CLI or REST API with token
REST API: Agent Resume
Resume suspended agents with new input via REST endpoint.
- POST /target/v1/agent/{runId}/resume
- Requires resumption token from suspension
- Optional input payload for new data
- Continues execution from suspension point
CLI: Agent Resume Command
Resume suspended agents from the command line.
- calljmp agent resume --target {RUN_ID} --resumption {TOKEN}
- --input flag for providing new input payload
- Real-time execution feedback
January 2026
Short-Term Memory Management
Automatic conversation history management for multi-turn agent interactions.
- memory.short.context(key) — create memory context
- Auto-load conversation history on LLM calls
- Auto-save responses to memory
- Seamless context passing without manual management
Prompt Studio
Manage and iterate on LLM prompts directly in the dashboard with instant updates.
- Create, edit, and version prompts in the dashboard
- Test prompts with sample inputs
- Evaluate against real traces from production
- Instant deployment — no agent redeployment needed
- Replay historical runs with updated prompts
calljmp typegen Command
Generate TypeScript type definitions for prompts and vault variables.
- calljmp typegen — generates .calljmp/types/agent.d.ts
- Type-safe prompt name access in code
- Type-safe vault variable access
- Auto-updates when prompts change
December 2025
Structured Output with Zod Schemas
Generate type-safe JSON responses from LLMs using Zod schema validation.
- llm.generate({ responseSchema: z.object(...) })
- Automatic JSON parsing and validation
- Type-safe response objects
- Fail fast on schema mismatch
Tool Calling API
Define TypeScript functions as tools that LLMs can call during generation.
- llm.tool({ name, description, parameters, execute })
- Zod schema for tool parameters
- LLM middleware calls tools automatically
- Support for multi-step function calls
Observability: Execution Traces
Full visibility into agent execution with detailed traces, phases, and logs.
- View complete run timeline with phase breakdowns
- See inputs and outputs for each step
- Inspect tool calls and LLM responses
- Error logs and stack traces for failures
- Duration and cost tracking per phase
Dataset Reranking & Reciprocal Rank Fusion
Improved RAG relevance scoring with advanced reranking algorithms.
- Reciprocal Rank Fusion for hybrid search
- Better ranking of multi-query results
- Configurable reranking strategies
- Improved semantic search accuracy
November 2025
Workflow Phases
Named execution steps with automatic status tracking and logging.
- workflow.phase(name, block) — define named steps
- Automatic timing and status updates
- Visible in Observability with step breakdown
- Parallel execution within a phase
Parallel Execution
Run multiple tasks concurrently with concurrency control.
- workflow.parallel(options, tasks) — run tasks in parallel
- Configurable concurrency limit
- Wait for all tasks or fail fast
- Automatic error handling per task
Retry Logic with Backoff
Automatically retry failed operations with configurable exponential backoff.
- workflow.retry(options, result) — wrap operations
- Configurable retry count (default 3)
- Exponential backoff between retries
- Customizable delay and backoff strategy
Datasets & RAG API
Semantic search over uploaded documents for retrieval-augmented generation.
- datasets.query(prompt, topK, minScore) — semantic search
- Returns ranked segments with relevance scores
- Metadata and source information included
- Configurable result limits and score thresholds
Datasets Upload & Management
Upload PDFs and documents for RAG ingestion directly from dashboard.
- Drag & drop PDF uploads in Datasets section
- 100 MB file size limit
- Automatic text extraction and embedding
- Progress tracking and status indicators
October 2025
Vault: Secrets Management CLI
Create and manage project secrets from the command line.
- calljmp vault add --name KEY --value SECRET
- calljmp vault list — view all vault entries
- calljmp vault delete --name KEY
- --sensitive flag marks values as encrypted
REST API Launch
Standalone agent invocation and polling via HTTP REST API.
- POST /target/v1/agent/run — start agent execution
- GET /target/v1/agent/{runId}/status — poll execution status
- Returns unique runId for async tracking
- Bearer token authentication with inv_ keys
Webhook Invocations
Configure webhook endpoints to trigger agents from external systems.
- Set webhook URLs per agent in dashboard
- POST to webhook triggers agent execution
- Request body becomes agent input
- Webhook responses include run ID for polling
Usage & Billing Dashboard
Track AI spend and usage metrics per agent and project.
- Per-agent usage statistics and costs
- Per-project aggregated spend
- Historical usage trends with date range filtering
- LLM invocations, tokens, and inference costs