Reading and context

Route	Returns
`POST /v1/context`	The context for one turn (see the options below)
`POST /v1/retrieve`	The evidence records themselves, best first, with `full_text`
`POST /v1/state`	The current value of something that changes (where someone works, a ticket’s status), with its history
`POST /v1/decisions` · `/v1/glossary` · `/v1/people` · `/v1/timeline`	Decisions with who and why · terms · people and roles · a subject’s events over time
`POST /v1/map` · `/v1/hierarchy` · `/v1/graph`	The code map · the organisation tree · the names graph (for the Inspector)
`POST /v1/memories/list` · `/v1/memory/view`	Browse and open records

/v1/context options:

Option	What it does	Trade-off
`query` or `messages`	The question, or the recent turns (the search reads them)	—
`max_tokens` (default 2500)	The memory budget	Fewer tokens can leave out evidence
`focus: true`	A shorter context: only the memories that match the question	It can leave out evidence the full context keeps. Use it for tight budgets, not for accuracy
`explain: true`	How each evidence record was found	Larger response
`include`	Extra fields: `evidence`, `map`, `memories`, `system_prompt`, `notes`	Larger response
`answer_format: "plain"`	Instructions for an answer without citations	Your model won’t cite ids
`viewer`	Reads as this person: their private records only (a user token is always its own viewer)	—
`filters`, `since_version`, `profile`	Narrow by tags or by what is new since a version; a stored ranking profile	—
`run_id`	An agent run’s context also cites the steps of similar successful past runs (coding agents: the past fix for the same code)	—
`used`, `outcome`	Which `[M…]` ids your model cited last turn and whether the user accepted the answer (ranking labels; no extra request)	—