Glossary

Alphabetical definitions of key terms used throughout Pencel.

---

Action Mode The agent operating mode where it can use tools, call external services, create artifacts, and take actions. This is the default mode for workflow steps. Compare with Chat Mode.

Agent A configured AI persona that performs tasks in your workspace. Each agent has a name, role instructions, a preferred model, and risk tolerance settings. Agents can work in chat conversations or execute workflow steps. See Agents.

Approval Gate A checkpoint in a workflow run where the agent pauses and waits for your manual approval before proceeding. Controlled by the Auto-Approve setting on each step.

Artifact A document, report, or file produced by an agent during a run or conversation. Artifacts are saved in your workspace and can be viewed, edited, or exported. See Artifacts.

Auto-Approve A setting on each workflow step that controls whether the step runs automatically or pauses for your approval. When turned on, the step executes without waiting. When turned off, the run pauses at an Approval Gate.

Briefing Card A card in the Pulse Feed summarizing a run's result — a generated headline, a short teaser, grounded highlights, and clickable source links — read in the Feed's list + reading pane and acted on with inline actions. Compare with Roll-up Card.

Chat Mode The agent operating mode for conversational interaction. In chat mode, the agent focuses on answering questions and having discussions without using external tools. Compare with Action Mode.

Connection A configured link between Pencel and an external service (e.g., Slack, PostgreSQL, Stripe). Connections provide agents with access to outside tools and data. See Connections.

Context File A workspace file that provides background information to agents. Context files are included in the agent's instructions when relevant, helping it understand your business, processes, or terminology.

Cron A scheduling format used to define when workflows run automatically. For example, "every Monday at 8 AM" or "daily at midnight." Set in the workflow's schedule field.

Decision Step A workflow step where the agent evaluates a condition and chooses which path to take next. Used for branching workflows (e.g., "if the value exceeds $100K, escalate; otherwise, proceed normally").

Episodic Memory A type of long-term memory that captures specific events or interactions. For example, "On March 1, the user said they prefer bullet-point summaries over paragraphs." Created automatically from conversation summaries. See Memory.

Guideline A rule or instruction that governs how agents behave. Guidelines can enforce tone, restrict certain actions, require specific output formats, or set business rules. They can be always active or activate only when relevant. See Guidelines.

Guided Task The "Your job" pane where you author a workflow — describe an outcome in plain language and Pencel drafts testable steps with you (per-step Test this step, Run once now), then Finalize to land it on the Jobs page. See Guided Task.

Hook An automation trigger that fires when a specific event occurs in your workspace. For example, a hook can start a workflow when an artifact is created or when a run completes. See Triggers & Schedules.

Jobs The run/operate surface (/jobs) — run history, live monitoring, a per-job trust rollup (clean / repaired / failed), and schedules. A workflow set up to run is a job. See Jobs.

Knowledge The unified surface (/knowledge) for everything agents draw on — documents, Context Files, Guidelines, Memory, and connected folder vaults — organized into a knowledge graph. See Knowledge.

Integration An external service that Pencel can connect to via the Model Context Protocol (MCP). Integrations include tools like Slack, GitHub, PostgreSQL, Brave Search, and others. See Integrations.

Long-Term Memory The collection of memories that persist across conversations and runs. Includes Episodic Memory, Pattern Memory, and Preference Memory. Agents draw on long-term memory to maintain consistency and learn from past interactions.

MCP (Model Context Protocol) The open standard that Pencel uses to connect with external tools and services. MCP provides a universal way for AI agents to interact with APIs, databases, and other systems without custom code.

Memory Information that agents retain across conversations and runs. Memory helps agents remember your preferences, past decisions, and important context. You can view, edit, and delete memories in Settings → Memory. See Memory.

Pattern Memory A type of long-term memory that captures recurring behaviors or preferences. For example, "The user always wants financial reports to include a comparison to the previous quarter." See Memory.

Plan Approval An agent setting that requires the agent to present its execution plan before taking action. When enabled, you review and approve the plan before the agent proceeds. This gives you visibility into the agent's intended approach.

Preference Memory A type of long-term memory that captures your stated preferences. For example, "The user prefers Slack notifications over email." See Memory.

Provenance The traceable origin of any output. Pencel tracks which agent created an artifact, which workflow step produced it, what data sources were used, and what model generated the content. Viewable in the run's event log.

Pulse Your home surface (/operations) for reading agent output and acting on it. Organized into scopes — Active (running / needs-you), Feed (result cards), and To-dos (personal checklist) — plus dynamic pack-record scopes. See Pulse.

Radar An opt-in background scanner that runs in a read-only mode and posts insight cards to Pulse — it observes and alerts but never takes action. See Radar.

Risk Tolerance An agent setting that controls how cautiously the agent operates. Low risk tolerance means the agent asks for clarification more often and avoids assumptions. High risk tolerance means the agent acts more independently.

Run A single execution of a workflow. Each run tracks its status, steps, outputs, costs, and a detailed event log. View runs on the Jobs page (open a job, then a run). See Runs.

Roll-up Card A single card in the Pulse Feed that collapses a record-producing run (e.g. an RFP scout) into one entry and routes you into its pack-record scope tab, where per-record actions live. Compare with Briefing Card.

Session A conversation thread between you and an agent in the chat panel. Each session maintains its own history and context. You can have multiple sessions open and switch between them.

Solution Pack An installable bundle in the Solutions marketplace (/solutions) that materializes a whole job — agents, workflows, guidelines, and managed connections — configured to work together, with its workflows compiled at install. Examples: RFP Scout, Stock Watch, QuickBooks. See Solutions.

Step A single unit of work within a workflow. Each step has instructions, an assigned agent, a type (action or decision), and auto-approve settings. Steps execute in sequence, and each step's output is available to subsequent steps.

Tailor The RFP proposal builder. From an opportunity card's Write proposal action, Tailor extracts requirements, drafts responses grounded in your past answers (from Knowledge and Memory), checks format compliance honestly, and exports a compliance matrix. See Tailor.

To-do An item on your personal checklist, the To-dos scope of Pulse. You or your agents create them; you can check them off or hand one to an agent as a one-off run (Add Task → Run once). See To-dos.

Trigger The event or condition that starts a workflow run. Triggers include manual execution (clicking "Run"), a cron schedule, a Heartbeat interval, or a hook responding to a workspace event.

Trust Grant A permission record that authorizes an agent to perform a specific action or access a specific resource. Trust grants provide fine-grained control over what agents can do.

Trust Tier A classification level that determines the default permissions for an agent's actions. Higher trust tiers allow more autonomous operation; lower tiers require more manual approval.

Workflow A sequence of steps that defines an automated process (formerly Playbook; the internal code and routes still use that term). Workflows can be triggered manually, on a schedule, or by hooks. Each step has instructions, an assigned agent, and approval settings. See Workflows.

Workspace Your Pencel environment containing all your agents, workflows, guidelines, connections, artifacts, and memories. Currently, each Pencel installation has one workspace. All data is stored locally on your machine.