Deliverd Context · knowledge, playbooks and objectives

Own your context.
And the record of what every agent was told.

Every agent starts from the agreed answer. Other tools give agents memory. Context gives them the agreed answer: the facts, conventions and decisions a person has confirmed, handed to every agent whatever model or tool it runs on. When something goes wrong you can see exactly what the agent was told and who said it was true. Playbooks write down how the job gets done, with the check each step needs before it acts, and objectives keep long work on track. All of it outside any one model, and exportable at any time.

Included in Team, Business Starter, Business and Enterprise · MCP, REST, CLI and SDKs for TypeScript and Python

Your context is yours: export the knowledge as Markdown or an AGENTS.md whenever you like, and import it back.

How the work runs

Step by step

  1. 01

    Write down what every agent should know

    Who approves supplier payments, which system is the source of truth, what we never do. Type it, import an AGENTS.md you already keep, or start from a kit.

  2. 02

    Agents propose what they learn

    An agent that finds out something useful proposes it with propose_knowledge. It is not given to anyone until a person confirms it.

  3. 03

    A person confirms it

    Confirmed words never change: a revision replaces an entry and the old words stay on the record. Each entry has an owner and a review date.

  4. 04

    Every agent is given the agreed answer

    Claude, ChatGPT, Copilot, Cursor and your own agents read the same knowledge through the same server, so nobody re-explains the business to each tool.

  5. 05

    Run the job the same way each time

    A playbook in force tells the agent each step and the check that comes first: a gate, an approval or a review.

  6. 06

    Answer "what was it told?"

    What each run was given, and who confirmed it, is recorded beside the work. So after the fact you can show what the agent knew when it acted.

What it does

Everything on this page ships today

Each part names what an agent calls for it and the plans that include it.

One knowledge base for every agent

Facts, conventions, decisions, contacts and known issues, searched by relevance. An agent proposes what it learns; a person confirms it before any other agent is given it, and a confirmed entry's words never change: a revision replaces it and the old words stay on the record. Secrets are refused.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
search_knowledgepropose_knowledge

What it was told, on the record

The confirmed entries relevant to a piece of work are handed to the agent with its objective, and what each run was given, and who confirmed it, is recorded. So you can answer what the agent knew when it acted, and who said it was true.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
get_task_context

Playbooks

A job written down as steps, each with the check that applies before it acts: a gate, an approval or a review. A person who runs the organisation puts it in force, or an approval of it does; running it starts an objective with a side task per step.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
get_playbookstart_playbook

Take it anywhere

Export the knowledge as Markdown that imports back, as JSON, or as an AGENTS.md for coding agents and other tools that read files. Bring an AGENTS.md or CLAUDE.md you already keep in, one proposed entry per heading.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
search_knowledge

A persistent objective

The outcome the work is for, kept apart from the conversation doing it. An agent can update the state and record findings; only a person changes the objective, so no model can redefine success and then meet it.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
create_taskupdate_task

A short task state

Where the work stands, the next action, open questions, constraints and what is blocking it. It is what an agent reads when it resumes after a restart, a new session or a different model, instead of the whole history.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
get_task_context

Side tasks

A side task starts from a snapshot of its parent — the objective, the decisions and the constraints — and finishes with a result: findings, proposed decisions, open questions and a suggested next action. They nest up to three levels below the objective.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
create_side_taskcomplete_task

Apply, keep or discard

Apply a finished side task to its parent and its decisions are confirmed in your name, its open questions are added and its next action becomes the parent's. Or keep it as reference, or discard it. Either way it stays on the record.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
apply_side_task

Findings and decisions

Hypotheses, tests, evidence, conclusions and risks, written for a reader to check — never the model's private reasoning, which is not stored. A decision an agent records is a suggestion until a person confirms it or approves it.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
add_findingrequest_decision_approval

The task's reports

Publish a report with the task's id and it appears on the task's page, and the next context the agent reads hands it back — so it revises the report it already wrote rather than starting another.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
publish_reportget_task_context

The task graph

Link work that depends on, blocks, supersedes or resulted from other work, and read it as one graph. Your systems hear when a task is completed or blocked, through signed webhooks.

Plans
Included in Team, Business Starter, Business and Enterprise
It calls
get_task_graph

Questions

Context and Deliverd.

How is this different from agent memory?

Memory is what an agent writes down for itself, and nobody checks it. Context is what your organisation has agreed: a person confirms each entry before any agent is given it, and what each agent was given is recorded. One agent's memory stays with that agent; Context is read by every agent you use.

What is a task?

A task keeps the goal of a piece of AI work apart from the conversation doing it: the objective, a short state — where it stands, the next action, the open questions — and side tasks that branch a problem and come back with a result. Any connected tool reads the same task, so switching models does not mean starting over. Objectives are included in Team, Business Starter, Business and Enterprise.

Which plans include Context?

Free includes a knowledge base of up to 25 entries, so you can see what agents do with it. An unlimited knowledge base, playbooks and objectives are included in Team, Business Starter, Business and Enterprise.

Can an agent write to the knowledge base?

It can propose. A person confirms an entry before any other agent is given it, and only a person retires one. Playbooks are the same: an agent may draft one; a person who runs the organisation, or an approval, puts it in force.

What stops us being locked in?

The knowledge exports as Markdown that imports back, as JSON, and as an AGENTS.md, from the app, the API, the SDKs and the CLI. It is read by any agent through the same MCP server, so changing model or tool does not mean starting over, and leaving Deliverd does not mean losing it.

Do we have to change the AI tools we use?

No. Objectives are reached through the same MCP server, REST API, SDKs and CLI as the rest of Deliverd, so ChatGPT, Claude, Cursor, Codex and your own agents read and write the same task.

Can an agent change what the work is for?

No. An agent can update the state, record findings and suggest decisions. Only a person changes the objective: an agent that could would be able to redefine success and then meet it.

Who can see a task?

Everyone in the organisation, unless the objective was started in a workspace — then that workspace's members and the organisation's administrators. A side task is exactly as visible as its objective. Owners, administrators, workspace administrators and publishers change tasks; viewers read them.

What happens to our tasks if we change plan?

Everything already recorded stays readable — on the page, over the API and to agents. Creating or changing tasks needs a plan that includes them.

Is the model's reasoning stored?

No. Findings are written for a person to check. The model's private reasoning is never shown or stored.

AI does the work. People make the calls. Deliverd keeps the record.