Can it remember why a prompt is worded a certain way?

Yes. Agents record durable observations, the failure that prompted a change, the wording that fixed it, the model quirk to account for, and read them back, so the reasoning behind a prompt persists instead of living in one person's notes.

Does the rest of my team see what I record?

Yes. Glen is org-scoped, so a finding one prompt engineer's agent records is readable by every teammate's agent in the same org, keeping the whole team aligned on hard-won lessons.

Shared memory for prompt engineers

Prompt engineers accumulate a quiet body of hard-won knowledge: which phrasings make a model reliable, which edge cases break a prompt, what tradeoff a given temperature buys you, which jailbreak a guardrail must catch, why a prompt was worded exactly the way it is. That knowledge usually lives in scattered notes, eval spreadsheets, and the engineer's own memory, and the AI agents you build never carry it. Each agent runs the prompt without knowing the dozen failures that shaped it, so regressions creep back and the same lesson gets relearned. Glen, shared memory for AI agents, gives your agents and your team one durable, shared memory exposed as a single MCP tool.

Prompt engineering is iterative knowledge work: you discover that a wording fails on a class of inputs, that an instruction must be phrased a certain way to hold, that a model version changed behavior you have to account for. Those findings are exactly the kind of durable fact that should persist and be shared, yet they typically evaporate into one person's notes. Glen makes them part of a shared store: an agent, or an evaluation harness, can read what the team already learned about a prompt, a model, or a failure mode before it runs, and write back new findings, so the institutional knowledge of why prompts look the way they do compounds instead of scattering.

For a prompt engineer this means the lessons behind your prompts become memory the whole team's agents consult rather than tribal knowledge you re-explain. The edge case someone found, the wording that fixed a regression, the model quirk that bit you, all become context an agent recalls when it runs or when a teammate iterates next. Because Glen is org-scoped, that knowledge is shared across everyone tuning prompts and across the agents themselves, so a finding one person records protects every agent in the org from the same failure. As a standard MCP server, Glen complements whatever eval and orchestration tooling you use and is readable from Claude Code, Cursor, or any other MCP client. Connect once over OAuth or an API key and let your prompt knowledge compound.

FAQ

Can it remember why a prompt is worded a certain way?: Yes. Agents record durable observations, the failure that prompted a change, the wording that fixed it, the model quirk to account for, and read them back, so the reasoning behind a prompt persists instead of living in one person's notes.
Does the rest of my team see what I record?: Yes. Glen is org-scoped, so a finding one prompt engineer's agent records is readable by every teammate's agent in the same org, keeping the whole team aligned on hard-won lessons.