--- title: Phasoric | Durable Project Memory for AI Coding Agents description: 'Give coding agents durable project context: architecture decisions, requirements, incidents and prior investigations. Connect authorized Phasoric knowledge through MCP.' canonical_url: https://phasoric.com/ai-coding-agent-memory markdown_url: https://phasoric.com/ai-coding-agent-memory.md --- Project continuity for coding agents # Your coding agent should know why the code looks like this. A fresh agent can read the repository. Give it the decisions, constraints and investigations that explain the code before it starts changing it. [Start with Premium — $14.99/mo](/signup?target=ai-coding-agent-memory&plan=premium&interval=month) [Find your client setup guide](/developers/mcp-client-setup?target=ai-coding-agent-memory) Premium includes 2,000 MCP operations and 25,000 hosted-intelligence credits per month. 1. Project notes / ADRs / research / decisions The history behind the repository 2. Phasoric durable context Only the knowledge you authorize 3. Codex / Claude Code / Cursor / Copilot Retrieve context → implement with client permissions 4. Reviewed decision / write-back Approve the note → better context for the next agent Ask for context before action ## The next task starts with the reasoning. > “Read the implementation brief and related decisions in my Phasoric Vault. Summarize the constraints and rejected approaches with source paths before changing code.” The brief, the repository and the agent stay distinct. Phasoric supplies authorized knowledge; your coding agent uses its own workspace permissions to implement the change. 1. 01 ### Keep the missing history Save architecture decisions, implementation briefs, rejected approaches, incident history, requirements, constraints, deployment decisions and known technical debt. 2. 02 ### Retrieve before implementing Ask the agent to find the relevant records and previous investigations. Check the source paths and unresolved questions before accepting a plan. 3. 03 ### Review the next decision After implementation, review a proposed decision note and approve its write-back separately. The next agent can retrieve that updated record. From our own work · September 2026 ## A plan in the Vault became a working change. We asked Codex to read a Phasoric Vault brief about persistent MCP connections. It retrieved the plan through Phasoric, investigated the OAuth flow, and implemented the fix in the repository. The reasoning traveled with the task. 1. 01 ### Read the brief An existing Markdown plan supplied the problem, constraints, and acceptance criteria. 2. 02 ### Investigate and implement Codex used the brief to guide the code changes and tests in the project. 3. 03 ### Review and merge The fix was merged into develop after approval, with release validation still explicit. Explore the workflow What did we decide?What changed?Write this decision back. Read the context. Do the work. Save the reasoning for next time. Recover the rationale Keep access tokens short-lived. Use rotating refresh tokens so an authorized agent can renew access silently. Keep vault-provider credentials behind Phasoric. Source: Persistent MCP OAuth brief · requirements and constraints The agent reads the original plan and its reasoning before proposing a change. Use the client’s own workflow ## A brief for Codex. A decision trail for Claude Code. ### Codex: start from the implementation brief Connect Phasoric on the host where the task runs. In a fresh task, ask Codex to read the brief, name its constraints and inspect the repository before editing. After review, propose the outcome as a separate decision-note update. Adding the URL and authenticating are separate steps. MCP knowledge access grants no repository or deployment authority. [Set up Codex](/developers/mcp-codex?target=ai-coding-agent-memory) ### Claude Code: recover the rejected approach Use the documented remote HTTP setup, then inspect and authenticate the connection with /mcp. Try: “Find the ADR for this subsystem. Which approach did we reject, and why? Cite the source before suggesting the same change again.” This is a suggested workflow, not a recorded Phasoric interoperability test. Claude’s remote connector and Claude Code use different setup paths. [Set up Claude Code](/developers/mcp-claude?target=ai-coding-agent-memory) [Prefer to ask Phasoric directly? Explore hosted intelligence](/hosted-ai-memory?target=ai-coding-agent-memory) Durable context. Deliberate access. ## Keep control of the record. Phasoric exposes the knowledge you authorize. It does not silently ingest every conversation or automatically remember everything your agents do. [Inspect the MCP boundaries](/mcp-knowledge-base?target=ai-coding-agent-memory) - ### Choose the source Hosted MCP reaches explicitly mapped Phasoric Hosted, Dropbox, Google Drive or GitHub vaults. Local-only vaults and device-encrypted remote bodies remain unavailable. - ### Keep the provenance Ask for the note path, evidence and limits of the answer. Your Markdown remains portable across tools. - ### Approve the write Review a proposed note change before application. Write permission, current-content checks and duplicate-request safeguards still apply. Documented client setup ## Start in the tool you already use. [ChatGPTBring project context into your conversations.Setup guide](/developers/mcp-chatgpt?target=ai-coding-agent-memory) [ClaudeUse a remote connector, or connect Claude Code.Setup guide](/developers/mcp-claude?target=ai-coding-agent-memory) [CodexRead the brief behind the implementation.Setup guide](/developers/mcp-codex?target=ai-coding-agent-memory) [CursorKeep engineering rationale beside your code.Setup guide](/developers/mcp-cursor?target=ai-coding-agent-memory) [VS Code / GitHub CopilotGive agent mode access to your project knowledge.Setup guide](/developers/mcp-vscode?target=ai-coding-agent-memory) An authenticated Vault read was verified in Codex. The other guides describe documented setup paths; client plans, versions and workspace policies determine availability. They do not establish Phasoric interoperability for every client version. [Prepare your vault and verify a first read](/developers/mcp-client-setup?target=ai-coding-agent-memory) ### More remote MCP setup paths OpenClaw, Hermes Agent and Grok document remote MCP connection paths. These guides map those instructions to Phasoric; authenticated interoperability has not been verified for these clients. [OpenClaw setup](/developers/mcp-openclaw?target=ai-coding-agent-memory)[Hermes Agent setup](/developers/mcp-hermes?target=ai-coding-agent-memory)[Grok setup](/developers/mcp-grok?target=ai-coding-agent-memory) ## Before you begin Does this replace repository instructions? No. Repository instructions still guide work in that checkout. Phasoric supplies the authorized project history, research and decisions that may live outside it. Ask the agent to retrieve that context explicitly. Does every coding session save itself? No. Phasoric does not silently ingest agent conversations or repository changes. Save the useful outcome and rationale as a reviewed note if you want a future task to retrieve it. What has been verified with Codex? An authenticated Phasoric Vault read supplied the persistent-MCP brief used in our real Codex investigation and implementation. This is evidence for that workflow, not proof of every client version or unattended deployment. Start with one project ## Give the next session something to build on. Keep the brief, recover the evidence and save the reviewed decision. Choose how you want to use that context. ### Use it through MCP Connect the AI tools you already use to your authorized Phasoric knowledge. Hosted MCP starts with Premium. $14.99/month includes 2,000 MCP operations and 25,000 hosted-intelligence credits each month. Higher plans increase limits and features. [Connect my AI tools](/signup?target=ai-coding-agent-memory&plan=premium&interval=month) ### Use hosted intelligence No MCP setup required. Ask Phasoric directly using hosted intelligence with bounded relevant source context. Eligible Free hosted accounts can buy 50,000 credits for $5.99. Usage is metered; a credit pack does not enable MCP. [Try hosted intelligence](/hosted-ai-memory?target=ai-coding-agent-memory) [AI agent memory](/ai-agent-memory?target=ai-coding-agent-memory) [Hosted AI memory](/hosted-ai-memory?target=ai-coding-agent-memory) [MCP memory & knowledge base](/mcp-knowledge-base?target=ai-coding-agent-memory) [AI memory for your Obsidian vault](/obsidian-ai-memory?target=ai-coding-agent-memory) [Context between AI tools](/ai-context-handoff?target=ai-coding-agent-memory) [AI projects & durable context](/compare/ai-project-memory?target=ai-coding-agent-memory) [Compare all plans](/pricing?target=ai-coding-agent-memory)