# Socra Cortex: Augment Your Agents with Your Knowledge > Give the agents you already use the knowledge they need before they act. Use Cortex when current, owner-governed context can change the correct decision and is not reliably available in your evidence. Canonical page: https://workspace.socra.com/solutions/agents Reviewed: 2026-08-30T08:43:39Z ## Agent contract - **CLI:** Flash one named Module. `socra cortex flash ` - **MCP:** Call context_flash with the target Module. `context_flash { "target": "" }` - **Returns:** The selected Module and its declared dependency closure. - **Then:** Verify current evidence before you change the system. ## Run this sequence for each task Cortex is pull-based. The agent chooses when to load the applicable knowledge. 1. **Read the Issue.** If an Issue exists, read it and take its Module name. Issue access does not grant Module access. 2. **Flash the Module.** Call Cortex explicitly for this task or session. Nothing is pushed into a later session. 3. **Verify current state.** Inspect the repository, runtime, tests, and external sources. Cortex does not inspect them for you. 4. **Act or expose the gap.** Follow the active instructions. If required knowledge is missing or conflicts, name the gap before you proceed. ## Know exactly what a flash does A flash assembles declared knowledge. It does not prove that the knowledge is complete or that the system follows it. - **Assembly:** For an acyclic closure, Cortex returns the target once with every transitively declared dependency, dependencies first. Current writes can store a cycle. A later flash fails. - **Access:** The caller needs access to the target. Its closure can include dependencies without separate direct access. Inclusion does not grant later access. - **Side effect:** Each success creates a durable, metered Flash record for the authenticated Account user. It does not change Module content or Issue content. - **Record:** The record stores its target, time, Account user, pinned Module revisions, closure size, and usage estimate: ceil(content characters / 4), summed once per Module. Wrapper text is excluded. - **Replay limit:** The record pins revisions, but reading its assembly again currently rebuilds the target's current closure. Save the original output when exact replay matters. - **Output:** The CLI prints the Flash ID, estimated token total, and Module count. MCP returns the assembled bundle as tool text. Both paths record estimated usage. ## Keep these boundaries active Cortex supplies one kind of evidence. It does not replace your authority or verification stack. - **Declared scope:** A closure contains only declared dependencies. Cortex does not find missing knowledge or semantic conflicts for you. - **Repository access:** Cortex sees only data sent through its API. It does not inspect your repository or runtime. Your agent host must do that. - **Repository routing:** For code work, the Issue must name the target repository. Cortex does not infer it from the Module, current directory, or repository name. - **Host precedence:** Cortex output is tool context. The agent host controls instruction precedence and tool confirmation. - **Approval:** Permissions control API access. Your team still defines and applies its approval rule. - **The Issue route is explicit:** An Issue records its Module. The agent still reads the Issue and calls flash. Assignment does not preload context. - **Instruction files remain:** Keep the instruction files your host discovers. They can apply globally or by directory. Cortex supplies knowledge that must be retrieved explicitly. ## Connect through the CLI or hosted MCP Use the path that is already available in the agent host. ### Socra CLI Requires a terminal, the Socra CLI, and an Account login. ```bash socra install cortex socra account login socra cortex map socra cortex flash ``` ### Hosted MCP Requires a remote HTTP MCP client that follows the resource_metadata URL in Cortex's 401 OAuth challenge, a Socra Cloud Project with Cortex enabled, and a current Cortex license. Test your host; use the CLI if it cannot connect. - **Server:** `https://cortex.socra.cloud/mcp` - **Flash:** `context_flash { "target": "" }` - **Identity:** A normal OAuth connection acts as the signed-in Account member. A separate Agent principal is used only when one is created and connected explicitly. - **Write surface:** The hosted MCP surface can write and delete. Account permissions still apply. Review its tools and your client's confirmation policy. ### A new Account starts empty Cortex seeds no Modules. Add one real decision to the smallest useful Module. Have its owner review the exact text before the first flash. `socra cortex module create --name --content-file ` ## Current cost - $20 per human seat each month after a 30-day trial. - Agents do not need seats. - Usage estimate: for each Module, ceil(content characters / 4); totals sum the closure. Wrapper text is excluded. - Reasonable first-party use is included. - Customer-directed third-party use is measured to the Project. Usage settlement is not active yet, so no usage overage is currently charged. - After a lapse, Cortex retains knowledge but blocks API, CLI, and MCP access. The current public contract has no export operation. Full pricing: https://workspace.socra.com/pricing ## Evidence status No public Cortex benchmark is published yet. Treat performance impact as unproven and run a controlled evaluation. Evaluate Cortex on repeated tasks with the model, settings, tools, task, and repository revision held constant. Compare final state, evidence quality, tool calls, tokens, latency, cost, and failures.